TLD
Code
TLD can be downloaded for testing. We provide a precompiled demo (Windows) and a source code.
A license has to be purchased for using TLD in closed source projects. The licencing is managed the University of Surrey: Commercial License
TLD is an award-winning, real-time algorithm for tracking of unknown objects in video streams. The object of interest is defined by a bounding box in a single frame. TLD simultaneously Tracks the object, Learns its appearance and Detects it whenever it appears in the video. The result is a real-time tracking that typically improves over time. Due to its learning abilities, TLD has been advertised under name Predator. The video to the right introduces Predator and demonstrates several potential applications. |
TLD was developed by Zdenek Kalal during his PhD thesis supervised by Krystian Mikolajczyk and Jiri Matas. The main contributions of TLD were presented at international computer-vision conferences. For his work on TLD, Zdenek Kalal was awarded the UK ICT Pioneers 2011.
More examples
Fast motion |
Recognition |
Similar objects |
Low-textured objects |
Key Features
- TLD tracks currently only a single object
- Input: video stream from single monocular camera, bounding box defining the object
- Output: object location in the stream, object detector
- Implementation: Matlab + C, single thread, no GPU
- No offline training stage
- Real-time performance on QVGA video stream
- Ported to Windows, Mac OS X and Linux
More Information
- High-level description of TLD
- Components of TLD
- Learning component of TLD
- Application of TLD tracker to faces
- Detailed description is in the following papers: TPAMI'11, ICCV'09 (w), CVPR'10, ICIP'10, ICPR'10
- Many technical questions (e.g. installation) are being discussed in the following discussion group.
Press
- University of Surrey -- Surrey student hailed as computer technology pioneer
- EPSRC -- Revolutionary approach to touch screen technology wins ICT pioneer award
- University of Surrey -- Zdenek Kalal wins ICT Pioneer Technology competition
- The Engineer -- Tracking system could help disabled people use computers
- Engadget -- Zdenek Kalal's object tracking algorithm learns on the fly, likely to make next 007 flick
- New Electronics -- ‘Revolutionary’ touch screen wins ICT pioneer award
- Gottabemobile -- Predator Object Tracking Algorithm the Future of Computer Interface?
- Physorg -- The Predator system helps the disabled to use computers
- Laptopmag -- New Learning Object Tracking System Called Predator is Amazing, Futuristic
- Tecmudo -- Estudante cria algoritmo para câmeras rastrearem objetos
- Hacker News -- Interesting discussion regarding Predator
- Reddit -- Interesting discussion regarding Predator
- Wired -- 'Predator' Smart Camera Locks Onto, Tracks Anything ... Mercilessly
- Popular Science -- Predator Camera Studies You, Tracks You Relentlessly
- Investorspot -- Can't Take My Eyes Off You - A Camera That Stalks You
- Time -- Revolutionary Object Tracking Video Software Released as Open Source
- New Scientist -- Smart camera learns to recognise you from any angle
- Slashdot -- Predator Outdoes Kinect At Object Recognition
- AIShack -- Predator: Tracking + Learning
- Geek -- Predator takes visual object tracking to new heights
FAQ
- How do you manage the number of templates in TLD?
For the moment, we do not put any constraints on the number of templates in memory. The number of templates typically stabilizes around 100, depending on the appearance variability of the object and background. - What kind of hardware is TLD running on?
TLD has been tested using standard hardware: webcam, Intel Core 2 Duo CPU 2.4 GHz, 2 GB RAM, no GPU processing is used and runs in a single thread. The demands of the algorithm depend on required accuracy of the algorithm. - To what degree is TLD invariant to light levels?
We use illumination-invariant features for detection. So as long as the changes in illumination are global, the performance does not change. If there is a strong local illumination change, the system is learning the appearance changes thus created.
(c) 2011 Zdenek Kalal