In part one of a two part series about tracking, Mike Seymour covers the history of 2D and 3D tracking in visual effects. Today’s artists might take it for granted, but what companies and individuals helped bring image processsing innovation to the field and change the industry forever? Read more… UPDATED: Aug 31st.
Tracking is the process of automatically locating a point or series of points from frame to frame in a sequence, allowing the user to stablise, track to or solve object or camera movement in the shot. The process started as one point tracking which could stablise a shot or add matching motion to a composite. Today it involves complex 3D camera solutions and extends to optical flow – the technology of tracking every pixel in a shot. In this series, we will explore the history of tracking, the best ways to shoot material for tracking, and provide an overview of current key products. Along the way we’ve interviewed some the men and women who invented the craft and many of the early adopters who showed its potential.
A VFX artist tracks or stablizes shots daily. Many also solve complex multi-point tracks and 3D camera solutions. But little is known of the inner algorithms that drive the software. By understanding exactly how the motion tracking works, it is possible to improve solutions and speed up work.
While it is increasingly true that almost any object or any scene can be tracked, this is our craft and profession. We should aim for the best results possible when on set and not allow the attitude of “fix it in post” to stop us from producing the best work possible. According to Doug Roble, Technical Academy Award winner for his TRACK software at Digital Domain, this is very important. “We go the other way,” he states. “We measure everything on set, we are really exact. 20% of the shots we solve are easy, 80% are hard. We focus on changing those odds”.
The art of tracking has changed with the advent of 3D tracking with tools such as 3D Equaliser, boujou, PixelFarm PFTrack, and RealViz MatchMover. It is changing again with floating point optical flow analysis from companies such as The Foundry and The Pixel Farm and emerging new laser scanning technologies. In the third part of the Art of Tracking series, we will take a brief look at these. While we will touch on Optical Flow Technology and Open EXR, this subject is so large they will be featuring in an upcoming Art of Optical Flow series.
Historical overview of tracking
Prior to digital tracking, in both electronic and optical effects, for most effects shots the camera was locked off. It was just impossible with out motion control of the capturing cameras to align to shots in post-production. While hand tracking has been attempted, the eye is so well adjusted to noticing movement and float – that often times sub pixel accuracy is required to sell a shot and fool an audience.
ln the beginning …
The US Defense Department first developed the concept of tracking for use in missile guidance systems. The earliest VFX use of tracking is perhaps in 1985 At the famous NYIT graphics lab Tom Brigham and J.P.Lewis (now at ESC) implemented an FFT based tracker that was used for a series of television commercials that we called the “rising coin” series. These were commercials for National Geographic where a gold coin traveled in an arc (as if it were a rising sun), and they had to be stabilized to match the movement and jitter of the background plates. ILM had an early 2D tracking software system called MM2 first tested on Hook and Death Becomes Her. MM2 was not an automated tool, but a manual 2d nudge tool where the artist would keyframe position changes by hand. ILM would later use this as the basis of one of the earliest 3D tracking systems which was used for Jurassic Park, development headed by Mike Natkin. Much of the early ILM work was done by J.P. Lewis now at USC’s Graphics Lab and Rod Bogart who joined ILM in 1995.
Lewis remembers, “At ILM I developed a Fourier-domain algorithm for *normalized* cross-correlation. Fourier cross correlation was a well known algorithm from the 60s (at least), but it was not known how to do the improved normalized version in the frequency domain. ILM let me publish the algorithm, and it was accepted to a vision converence, Vision Interface 95. A version of the paper is online at http://www.idiom.com/~zilla/Work/nvisionInterface/nip.pdf
Since then this algorithm has spread around, including now being implemented in the Matlab image processing toolbox, etc. Also, I separately reimplemented it, and contributed the algorithm to a couple of packages including Shake and Commotion.”
Joe Alter developed one of the earliest markerless tracking systems “tga_stablize” on an IBM PCAT(286) with Truevision Targa 32 card to match two morph plates for Star Trek, The Next Generation in early 1991. “Later that year I re-wrote it for the SGI PI-30 and the IBM-PVS parallel super computer at Boss Film for use in Cliffhanger,” explains Alter. “I was the a lead TD along with Andy White, under CG Supervisor Jim Rygel (now at WETA). The utils were called “rle_coarse” (for pixel level tracking), “rle_fine” for subpixel tracking, and ‘rle_track’ (adapted from Graphics Gems witness point tracking util) for finding 3d camera positions from point data.”
On the 1993 feature film Cliffhanger, Andy White took the tracking data and fed it into textured polygon corners in wavefront TAV to cover up holes in the set construction (a fake cliff with 2 by 4s sticking out), and Alter used it to reconstruct clean plates from moving cameras for rig removal and tracking of the live plates into various matte paintings. Boss Film did excellent work, but Alter would later move on to join the ILM’s team in ’94.
It was Discreet who first brought tracking to the broader visual effects community in their landmark Flame system. Discreet lead the field for many years with their innovative and brilliant tracking developments. Emmy-winner Wally Rodriguez, of Miami visual effects studio Upstairs comments, “It (tracking) was in the first version we got of flame and depending on who you talk to Jake Parker and I were either delivered second or third systems. (The very first non-feature film system was delivered in the US to a place called “Walk the Bible Productions” in Atlanta). – Tracking and timewarping were the big sellers for us.” .
LA flame artist Maribeth Emigh relates, “For the first version I had of flame (and I was one of the first past the mario flames [ed. Mario Brothers feature film unit]), I remember a 1.0 scribbled on the exabyte I had. It had 1 point tracking…and it worked rather well but limited for just one point. You looked for the highest contrast and could only do frames and forward…. but at the time the fact that you could track at all saved tons of time.”
Flame started life as Flash – written in the suburbs of Melbourne Australia by Gary Tregaskis. Tregaskis, who has long since moved away from Flame and Discreet wrote Flash and it moved to Canada and the newly formed Discreet Logic. Flame was first shown at NAB 1992 and offically launched at Sigrapph 1992.
Ken Deaton wrote in the June 1992 issue of Video Innovations “You’re afraid to ask what’s next – I can tell you still think your black box can beat my Eddie then you should know that Eddie has a big brother, Flame. Flame is also brought to you by the folks at Discreet Logic. If you can’t get your head around all that jazz that EDDIE does, try imagine EDDIE on steroids with a near-real-time attitude. That’s because FLAME utilizes the aggressive graphics processing power of SGI VGX computers’. Big Box. Rather expensive one too. It helps to have an Abekas recorder too (doesn’t it always). But if you have the funds,then you can have some fun doing that morphing thing, right up to IMAX resolution. I would tell you more about it, but I can’t afford it, anyway. Besides it’s just in beta now and doesn’t officially debut until SIGGRAPH this summer. Remember you heard it here first”.
In an exclusive, Tregaskis spoke to fxguide about the moment he invented tracking for flame, “Tracking… Hmmm… Well, I first thought of tracking way way back in the early 90′s… I was at a facility in LA… Rich Thorn’s facility… Can’t remember the name of it..(ed. Digital Magic), while watching a Harry operator working on a shot of a car falling over a cliff… He was manually trying to composite a shape over one of the windows of the car to obscure what was inside (or the fact that no one was inside)… It was remarkably unsuccessful as the shot was shaking all over the place and the angle of the window kept changing… I thought there had to be a better way… So went home and wrote it… It was back in the days when CPUs ran as fast as a washing machine processor… So the only way to get it to run at any decent speed was to pump the image data through the accumulation buffer… Using the accumulation buffer to perform mathematical operations on each pixel simultaneously… It worked well for a first implementation… But was subsequently blown away by much faster cpu’s and brilliant coding by some of the brains at discreet…… I remember the day clearly.”
Peter Webb, who is credited as the world’s first flame artist, started along side Tregaskis in Melbourne. “I do remember when Gary showed me. He was always excited to bring a new alpha build and would watch for the ‘WOW’ reaction when he showed us new stuff like displacement with lighting. The tracker was one of those moments. We had been matching plates by hand which was insanely laborious and to see the software rapidly chugging through the shot with that little red box locking on perfectly was mind blowing. That was part of the wow factor. The rest was the realization of how incredibly useful this feature was.”
The first system used for film use was in 1992 when flame started work in the feature film Super Mario Brothers. Super Mario did not make the Oscar nominations, but it was included in the bake off that year. By the end of 1992, Flame had moved into commerical work at Lamb and Co. in Minneapolis, Minnesota.
In 1993 Flame started to be noticed and Version 3 released.
The early hardware config required to run flame and its new trackers is:
- Silicon graphics VGX 420
- with multi-Buffer and FX option
- 64 Megabytes of Memory
- Standard High Density Tape Drive
- 2Elite2 Fast SCSI-2 Drives with controller card
- Videolab card
- Abekas A66 Digital Disk Recorder
- Wacom 3D312E Pressure Sensitive tablet
In 1994 the first flame Siggraph users group meeting met in Orlando. V 3.9.10 was released and it was the last version for VGX graphics, from then on Discreet supported the Onyx range of SGI computers. I/O was not part of the inital product so early artists had to use an Abekas to bring in and lay off to tape.
” I remember demoing the tracker myself in the early 90′s at trade shows and we always used this footage of some Russian soldiers marching,” Webb tells. “We would track the Russian emblem on their fur hats and then replace it with a flame fish. The punters LOVED it as it had never been seen before. Single point tracking first, there wasn’t any multi point tracking in the first version”, he laughs, “the reaction was always ‘how the fuck are you doing that ?’”
Rodriguez remembers “that when we were making the decision to buy, we had traveled to California to see the people from Colorgraphics about Composium. I made a list of stuff that we liked and that I thought should be on flame. Russell [ e: Russell Weaver, one of the original investors in discreet] promised that if in one year all of this stuff wasn’t in the system they would refund us 110% of the cost of the system. I remember that Gary used to travel from site to site, fixing site-specific things, and configurations that might not be the same as at discreet. He would then take his tar tape and go to the next site. They made their 1 year deadline.”
It would be several years before any other product would allow tracking for VFX work. Webb also remembers ” quantel coming out with some tracking software sometime after discreet. Trying to pretend they had thought of it years ago… Ah… The dog stab dog world of image processing in those days… Now its all so samey and runs on my laptop…if imitation is the sincerest form of flattery, Gary’s tracker should be blushing”
Tracking was first shown on a Quantel Henry in 1994. The demo involved tracking a Quantel logo onto an elephant. According to Quantel’s Steve Owen, “Some of the quantel team had worked on image tracking years earlier in a lab but in a pure research capacity.”
Colin Wrey did the first implementation – called ALF for Auto Lock Follow. Wrey is one of Quantel’s resident geniuses and is still at Quantel working on the new generationQ product. “Once we decided to do it the basics were simple – it took a couple of days for the first implementation – the trick was to make it useable,” remembers Wrey. “There’s a balance between a perfect track that takes forever and an interactive track that falls off from time to time. We decided to go with interactivity and the real thought went into making it easy to use, making it easy to stop and restart a track because it was never going to be perfect.” Since the early days of development 4 point tracking, the ability to track outside the picture area, offset track for dealing with track points that are obscured for part of the path, and graphical manipulation of tracking data were all added to the Henry .
The iQ was first shown at IBC 2000. iQ (and eQ and gQ) have a completely new tracker with up to 99 points, including on screen manipulation of these points. At the moment, Quantel has no 3D camera tracking.
In 1995 there was a major advance in tracking as Discreet moved from version 3 to version 4. At the second flame user group meeting, head of development Terry made his famous quote now known by old timers as simply “Terry’s famous quote.” “We will now always ship 90 days from show,” said Terry. In reality it was nearly 12 months before the new software and new tracker would be released, but with it came a major improvement in tracking. Behind a large part of that improvement was Benoit Sevigny.
Visual effects Supervisor Sheena Duggal explains that, “the original flame had a one point tracker,…seemed to work by doing some kind of emboss between frames and required a point with maximum contrast. When Discreet bought flame, the tracker was updated by Andre Le Blanc (one of the original founders). At this point it became a 2 point tracker and if I remember correctly you could track scale and rotation. Then a year or two later Beniot made it even better.”
“I do remember going to discreet and talking to the boys and Benoit” says Emigh. “They were an amazing bunch…the way they thought this stuff up as if it was a ‘hey look what I fooled around with this morning’…. I do remember tracking being one of the things that made flame stand out above Q”.
At this time Mike Hughes was in charge of R&D at Discreet. “Benoit worked on just about everything, including Soft Edits, at one point or another,” says Hughes. “I was actually coding the early scratch track audio alongside Benoit before I took over managing Flame. At the time of the v3-4 transition, I’d estimate that there were about 15-20 developers working on the release, which included the video and s w people directly implicated.”
Sevigny started programming Fourier transforms at the age of 11 and spent more than 12 years optimising it before he was allowed to deploy his solutions in Flame’s tracker. After joining Discreet he was initially assigned to working on OMF I/O and even though he was an digital signal processing expert he designed Discreet’s database file format.
Inspired by mathmatical legend Jean-Baptiste Fourier and years of writing real-time processing/analysis on a Commodore 64, Sevigny set to dramatically improve the speed and accuracy of tracking. “The pre-4.0 Stabilizer was originally designed by Gary Tregaskis and was maintained for a while by Andr’ LeBlanc,” says Sevigny. “It used a least-mean-difference method implemented in a brute force manner using the graphics hardware. Its original purpose was to stabilize unsteady shots but the integration in Action as a general tracking tool really allowed to make most of the technology. But the main problem was that in the early days, tracking a 30 seconds spot was a half an hour batch process that was not guaranteed to succeed. Although the idea of providing a tracking tool in an online compositing system was at the time revolutionary, it was prohibitive to lock down an Onyx-class machine for this task and to supply coffee during the many breaks the artist was forced to take in between retries.”
“When,” Sevigny remembers, “in a wild move, Michael Hughes offered me to work on the tracker, the tool had a very basic UI and supported up to four trackers when accessed through Action. I’m quite sure that at the time, nobody expected me to transform the tool in such a radical way.” For the Flame 4.0 Stabilizer, discreet and Sevigny mainly wanted to bring tracking to an interactive level, that is, on par with the rest of the application. “In this respect, I think my most important contribution is the design of the first widely available truly interactive tracker and this gave Discreet’s products an edge over others for years to come” comments Sevigny. “This first tracker of my own used a cross-correlation method using Fast Fourier Transforms (FFTs). This method is indeed fast but is not in general fool-proof. To help it in some circumstances, I provided two analysis modes: a luminance and an edge mode. The luminance mode used a standard cross-correlation while the edge mode relied on phase correlation. The three colour channels were processed independently and the final match was a combination of the three results”.
In version 4.0, Discreet also introduced a parameter that allowed to reject results that were below some correlation threshold, thus keeping the tracker from being fooled by objects in the foreground. This was demonstrated for the first time to ecstatic users at the 1995 NAB user group meeting in the now famous Speed shot where a bus is tracked as it passes under a bridge and momentarily out of site. At the time “it really seemed like magic to us” jokes Webb.
Also new in 4.0 was the patented reference switch mechanism that allowed to switch to another feature when the first one is occluded of goes out of screen. Sevigny recalls, “One day, while I was working on grain management at the London office, Sean Broughton from Smoke ‘n’ Mirrors came by and challenged to track a sequence he had a hard time with (if I remember well, it was the Levi’s “Planet” spot). The shot, with an ever changing moving feature, could partially be tracked with Fixed Reference on, but with the roaming reference mode, the error would build up to an unacceptable level. The only workaround was to track successive segments with different reference frames (in fixed mode) and to try to stitch together the animation segments, which is far from being an easy and accurate task. I then realized that when the fixed reference mode is on, the reference should be updatable on-demand (instead of continuously with fixed reference mode off). This would give the highest accuracy while letting the user decide when the reference should be updated. Thanks to Sean, the Snap function appeared in the following release. To use it, simply back off the timeline just before the tracker looses track and snap the reference. Then restart the analysis from that point. Snaps are remembered in the reference Channels, so reanalyzing the sequence will automatically make use of them.”
Jerome Chen, Gary Jackemuk and Ron Brinkman gave a class at Siggraph on compositing. From this Brinkman would go on to write the now definitive reference book ‘The Art and Science of Digital Compositing’. Brinkman, formerly of Sony Pictures Imageworks already had a long string of credits including In the Line of Fire, Die Hard: With a Vengeance, and Speed . “After finishing up the work on Zemeckis’ ‘Contact’ I decided that it was a time to take a break from production and try something different,” says Brinkman. He become one of the original people behind Shake and Nothing Real and is now at Apple. Shake — originally planned to be named IT (for Image tool) or even “Look” — took the film world by storm. This was primarily due to the fact that it worked in a higher bit depth than anything else on the market, pushing past Inferno’s 12 bit linear to floating point, yet also running on desktop machines.
“An early version of Shake (possibly before it was called that, and possibly still in a prototype shape) was actually used on the opening shot from Titanic,” says Brinkman. “A deep underwater shot where everything was a similar shade of ocean- blue. The shot originally been started at 8 bits/channel of colour information but the subtle tonal gradiations brought out tremendous banding problems. Shake was brought in and its higher bit-depth capabilities were crucial to getting a clean artifact free shot out the door”.
Meanwhile, Discreet had not stopped working on tracking and at their NAB 96 user group a new Flame 5.0 was previewed for the first time with another complete overhaul of the algorithm. “While version 4.0 had set the pace for the version to come, I was not totally happy with it. Some people preferred the pre-4.0 because it was more forgiving. This is why, after a digression working on grain management tools, I decided to make a come back and revisit the whole approach.” recalls Sevigny. “This time, I wanted to fix some accuracy and robustness problems while again improving on the speed. The definitive algorithm does not use cross-correlation at all (although some people pretended it does). It is more pre-4.0 like, but it still uses my beloved FFT implementation to efficiently do the computation. This method is so robust that I didn’t need to provide distinct modes for luminance and edges anymore, so it worked out of the box. After extensive tests, I also realized that trying to combine different results from three colour channels was a source of instability, so I decided to merge the colour channels upstream, thus making the tracker colour-blind as a side effect. It proved to be more robust to illumination changes and required three times less processing”.
Rich Bobo, Senior Flame artist, then a demo artist for Discreet recalls “Yes. Benoit was a clever, young, mad scientist type of programmer – hopped up on espresso – and full of great ideas. I remember him showing me all these tracking tests he was doing, using FFTs and other types of pixel comparisons. He experimented with center weighting of pixel grids, separate R, G and B tests, luminance weighting, etc. At one point, there were a lot of individual controls that the user had to play with – and, it was not very intuitive. His thought was that it would be more flexible, to allow tracking in different situations. However, small tweaks could have very different results and it was very fiddly. I kind of gently suggested that what the user would want would be a “one-button” solution that would “just work”. He looked rather dismayed but, to his great credit, he spent a ton of hours trying to make an all-in-one version that would be a no brainer tracking solution. In the end, he came up with exactly that – and, he made it really, really fast! I recall him proudly stating that the tracking algorithm could do analysis at 70 frames a second, if it didn’t have to wait to load individual frames from the disk array!”
Sevigny recalls “actually, I wanted from day one to provide a “one-button” solution but it took some time before a I solve this difficult problem.” Being a musician myself, I always thought that ultimately artists should not have to fiddle with numerous and obscure parameters to have to job done.” But when it’s necessary because the algorithm doesn’t handle all cases, it’s nice to have them. Finally, 5.0 offered what I had in mind from the very beginning”
The idea behind the increase in speed was not about showcasing how much faster than real-time the programmers can get, but in making more demanding uses of tracking practical. Previous use of the tracker in flame required either 1 tracker (stabilization), 2-3 trackers (translation/rotation/scaling) or 4 trackers (corner pinning). In order to tackle new applications like garbage mask and warper mesh tracking, the new algorithm had to be lighting fast and had to support an unlimited number of trackers. When tracking hundreds of features, the slightest difference in speed is magnified a hundred fold. Another important issue with speed was the advent of high resolution digital film and HDTV around this time. With a larger-sized reference and tracker area, much more processing was required.
“Tracking algorithm design being a black art (it’s nearly impossible to make something that will work in all situations), I also wanted the users to easily correct inaccuracies.” says Sevigny. “For that matter, I developed a totally new viewer in which you could edit motion paths in context through a patented overlay which allows you to manual register the reference frame to the current frame. Managing hundreds of trackers can be a daunting task and I hope this new viewer made this easier.
Version 5.0 of Flame shipped in 1997. After version 5.0 of Flame, Sevigny left Discreet and now works at Kaydara doing real-time character animation including automated lip syncing. Around this time, Australian Steve Roberts was moved to Canada. Roberts had been part of production company NYPD in Crows Nest Sydney. Digital Fusion was a proprietary compositing solution at the post facility Steve started in Sydney. Shortly after some serious interest by Toronto based DPS, eyeon was born and moved to Ontario, Canada in early 1997 where it is currently based. Rony Soussan, was one of the first ever Digital Fusion users, Rony Soussan, ” I was one of the first customers to purchase Digital Fusion and I can tell you that it has gone through quite a few changes. It first appeared as a single point tracker, then eventually a 4 point corner positioner. Soon after that we began building triangulation into the tracker and we created a stabilizer version. Now, you can track as many points as you want, have automatic triangulation, stabilization (see tip below), corner tracking, perspective tracking, and you can even embed a tracker into any animatible point or control within Fusion. The accuracy has also become significantly better and faster. I can quite honestly say without question it has become one of the best, if not the best tracker in the industry”.
Eyeon’s Digital Fusion was for a time bundled with Alias. Eyeon did two flavors of Maya Fusion for Alias. A light version shipped with all versions of Maya for Windows. There was also Maya Fusion, which was essentially the same as Digital Fusion except for having Maya marking menus and looked the same as Maya. It stopped in Feb 2001 when Alias consolidated its focus into Maya 3D. At the same time they stopped the A|W Composer development as well
The product was significant not only for providing 2D tracking tools to the PC market and then evolving them into days powerful cross platform tools with full floating point support and the world’s first Open EXR support. Alias also released Maya Live with the Maya Unlimited set. Surprisingly, Maya Live was developed just down the road from NYPD in Sydney’s Crows Nest by post-production company Garner McLennan Design (GMD). GMD eventually sold the software to SGI/Alias wavefront. Chris Horbath, then in R&D at GMD and now at ILM, started work on a software colour correction package. He needed a framework for it to side within so he started developing Raindance, a product not unlike Flame. As part of this basic work he started working on a colour window tracker to allow colour corrections to track shapes in the image. He then hit upon the idea of taking this technology and using it to dynamically track and adjust two shots to align with each other. to do this he needed to triangulate the camera. GMD’s Rob Nicol and Simon Brewster took the product to NAB 1996 and sold it to Alias, it appeared first as a 2d tracker in composer but this was right when Alias was merging with Wavefront and looking to introduce Maya, the 3D tracker would appear as Maya Live a year later .
1998 was the year tracking started to be acknowledged by the industry awards. At the 1998 Oscars, the Scientific and Technical Academy Award went to Gary Tregaskis for the primary design and to Dominique Boisvert, Phillippe Panzini and Andre LeBlanc for the development and implementation of the Flame and Inferno software
Also that year Dr. Douglas R. Roble won the technical Achievement Award for his contribution to tracking technology and for the design and implementation of the TRACK system for camera position calculation and scene reconstruction. The TRACK system is an integrated software tool at Digital Domain that uses computer-vision techniques to extract critical 2D and 3D information about a scene and the camera used to film it. Thaddeus Beier of Hammerhead won for the design and implementation of ras_track, a system for 2D tracking, stabilization and 3D camera and object tracking. Hammerhead productions was founded in 1995 by Jamie Dixon, Thad Beier, Rebecca Marie and Dan Chuba — all ex-Pacific Data Images employees. PDI was then a ground breaking effects house and the first company to invest in workstations over massive main frames and had a vast history of impressive work. This was before the Dreamworks buyout of PDI and its later Shrek fame.
imright(art_of_tracking/track_3De4.jpg)While Roble won the Technical Achievement Award in 1998, the first version of TRACK had been developed in 1993. TRACK, now in version 5, has been completely rewritten and is still used today at Digital Domain. TRACK feeds Digital Domain’s (d2) NUKE compositing system with full 3D camera solutions (Nuke would itself win a Technical Achievement Award 3 years later in 2001). Roble is the first to admit that tracking is still “very much an art”. He sees the role of his team which includes John Flynn and Henrik Falt as providing the artists at Digital Domain with a range of tools that allows them to solve very complex problems. He explains that artists are often required to start with one method, then use another and finally a third or fourth approach to solve many of the more complex problems they face. TRACK is therefore a very flexible tool allowing a number of different ways to track and solve a camera move. The software has moved from its roots as a 2D tracker to now encompassing Optical Flow and a highly complex 3D camera tracker.
The move of TRACK from 2D to 3D tracking was reflected by a general shift in the industry to 3D. Just the year before in 1997, Science-D-Visions in Dortmund, Germany became the first software vendor to release a reliable survey-free 3D camera-tracking application. Since then, 3D-Equalizer has become one of the key matchmoving systems of choice for the digital media industry.
In early 1998 REALVIZ was founded with the goal of commercializing products based on over ten years of computer vision and robotics research by INRIA, the French National Institute for Research in Computer Science and Control. The company founders are former executives from Medialab, a subsidiary of the French TV channel “Canal “, and former research scientists from INRIA.
REALVIZ successfully launched their first four applications ImageModeler, ReTimer, Stitcher and MatchMover by the year 2000. These were part of the next generation of tracking programs — those that aimed to solve 3D tracks and produce an exact camera move in 3D space that could be used by 3D and 2D programs alike. By the start of the new millennium, tracking R&D battles had moved from the big iron boxes of Henry and Inferno to dedicated PC programs harnessing the power of the desktop and the latest in University R&D.
“The main benefit from matchmoving software,” says Luc Robert , Chief Technology Officer at REALVIZ, “is a marked increase in productivity. Automatic matchmoving is definitely a huge help in this respect, since on a significant fraction of shots it produces a solution which requires no human interaction at all. Most post-production companies using automatic matchmoving have integrated it into their pipeline so that, for virtually no overhead cost, they can benefit from these automatic solutions. Despite this, there will always be shots which automatic matchmoving software will not be able to solve (the extreme case being where there is absolutely no element visible in the footage on which the tracking algorithms can rely). It is crucial for professionals to solve 100% of the shots they want to matchmove, and for that, the software has to allow them to control and guide the process if necessary.” REALVIZ proved very successful over the next 4 years, MatchMover technology has been used to create visual effects in films such as ‘Troy’ , ‘The Last Samurai’, ‘Hellboy’, ‘Daredevil’ ‘Dinotopia’ and Harry Potter
The battle for technological leadership would be fought by REALVIZ, 2d3′s boujou and the innovations at Science .D. Visions in Germany with their 3D-Equalizer software, who would themselves win a Technical Achievement Award in 2001 for for the development of “3D-Equalizer , an advanced and robust camera and object match-moving system.” which in the words of the Academy, “this commercial tracking system provides “survey-free” tracking, which significantly reduces the need for painstaking, error-prone measurements on sets.”
3D-Equalizer’s feature film credits are impressive. Lord of the Rings (all three), I, Robot, X-men 2, Spiderman, Gladiator, MI-2 and many others including the Emmy Award winning Walking with Dinosaurs. At the end of 2000, 3D-Equalizer maintained a technical lead in the industry with such innovations as solving footage using zoom lens and allowing secondary images to be used to help find a solution. In 2001, 2d3′s boujou was released.
2001 – 2002
boujou was launched at NAB 2001, using advanced adaptive algorithms developed from vision science research. The software greatly simplified what was achievable by normal 3D professionals and the product took off. So much so that in various markets the product name became a verb. “We’ll boujou it” was heard on sets and in studios all over the world
boujou has made a major contribution to high-profile television and film productions, music videos and commercials by facilitating the creation of visual effects that seamlessly combine live action and 3D. In 2002, 2d3 was awarded with a Primetime Emmy Engineering Award for boujou. 2d3, based in Oxford, is a wholly-owned subsidiary of OMG plc which also owns Vicon Motion Systems, developers extremely advanced motion capture systems. boujou has been used on such films as The Matrix series, Harry Potter, Lord of the Rings and Troy. Playing on their strengths, 2d3 released boujou bullet in 2004 – an even simplier version of their boujou two software. Bullet offers all the functionality of version two with a new built in ‘wizard’ feature, all for a fraction of the price of the full version. Version three of the main product will ship in September 2004 with new features such as solving zoom lens and a new automatic radical distoriton model. Lens distortion is a major issue for tracking programs, but according to 2d3′s Steve Hill, “We’re going to supply a Shake node with boujou3 that will allow you to use the boujou lens distortion model to add or remove distortion ” Shake already has a distort and most importantly – invert distort function, but the new plugin for shake is a special lens distortion node. Shake distortion nodes are extremely important to many pipelines, REALVIZ’s Liz Tjostolvsen points out that by the end of September 2002, MatchMakerPro 2.5 was released with just such a Shake distortion node.
Today, new products appear regularly and prices have fallen. Oscar nominated Visual Effects Supervisor Nathan McGuinness joked at a recent AEAF conference that Asylum used a whole range of tracking tools on the incredible task of tracking ships and oceans for the film Master and Commander. This included Andersson Technologies SynthEyes 3-D tracker which retails for only US$349. Andersson Technologies LLC was founded by Russ Andersson in August 2003 to commercialize SynthEyes. Andersson had previously been at Bell Labs for 15years and also wrote SceneGenie. SynthEyes was first marketed in December 2003, at which time it had already been used by beta testers in feature films such as Bad Boys 2, Charlie’s Angels 2, and Master and Commander: Far Side of the World. It is an amazing price point but also very fast. Andersson comments that “it has been designed from the ground up to be fast. There are a variety of techniques used internally to make this happen. Sometimes a bit of planning can save quite a lot of time, even for a computer! ” He later jokes “I have a background in robotics and real-time computer vision, where speed is the norm.”.
Another interesting product came from the public domain. PFTrack was developed using a licensed technology base from a product called Icarus. Product Director, Mike lancaster came across Icarus while looking for key technology to incorporate into Pixel Farm’s as yet un-announced long term strategy. The Pixel Farm approached AIG and quickly negotiated an exclusive license deal to utilize technology from Icarus because the support and management of the software was causing untold problems for AIG. Icarus was distributed free to non-commercial sites, with a commercial license available for use within professional facilities. According to Lancaster, “you can count on 1 hand the number of commercial licenses sold, yet there were some very big names on the IP listing which was testament to the quality of the product.”. In many respects, The Pixel Farm saved Icarus, developing a fully professional application from the base technology. The Pixel Farm has created a complex analysis tool capable of far more than just match moving designed around complex long form projects. To address the match move market, The Pixel Farm spun off PFMatch as a low cost desktop application with a full set of camera tracking tools but lacking the high end tools such as batch processing, proxy resolution handling and application plug-in support found in PFTrack.
The world’s leading desktop effects solution, After Effects (AE) from Adobe, also includes a new tracker. AE has placed the technology in more hands than perhaps any other single product, providing a huge group of people the chance to use tracking daily. While AE does not do 3D tracking, Adobe’s After Effects Michael Coleman, comments “that Boujou has a plug-in for importing their data to After Effects. Also, AE can import camera data from Maya .ma files, so any 3D tracker that can create a .ma file can apply the 3D camera data to an AE camera. AE can also extract camera data embedded in RLA sequences, which is of use to certain 3d people.” AE has advanced 4 point tracking that is both impressive and very useful. “An advanced capability of our perspective corner pin tracker is it’s ability to apply the surface to a different surface than was tracked,” Coleman states. “In other words, if you track four points on a wall, you can correctly corner pin something to the floor plane, regardless of what’s on the floor”.
AE was initially not incredibly fast at tracking but Adobe has done a lot of work in speeding it up. “We have put significant effort into the tracker over the past two releases.” explains Coleman. “The list of improvements is long, but top of the list must be accuracy and speed. It can be 20-50 times faster in some cases. All of the track data is fully integrated into the AE project and available on the timeline for parenting, expressions, or simply copy and pasting tracker keyframes between layers. Users have reported excellent accuracy results recently, particularly with After Effects 6.5. Also new to 6.5 are is one-dimensional tracking, 2-point scale tracking, zooming the track point when you are dragging it, and showing the motion path for the resulting track data.”
In the next section we’ll discuss what works best on set, how to improve your tracking solutions and summarize, available 2D and 3D applications.
We've been a free service since 1999 and now rely on the generous contributions of readers like you. If you'd like to help support our work, please join the hundreds of others and become an fxinsider member.