Realtime Object tracking systems
A picture speaks a thousand words, videos tens of thousands. Of all the senses man is endowed with, it is unquestionably the sense of vision that feeds the brain with the most amount of information. Not surprising therefore, is the amount of interest evinced in the field of computer vision by researchers all over the world. Despite promising developments taking place by day, one of the biggest challenges impeding an even more rapid progress in the field of computer vision is the challenge of processing in real-time, the “ten of thousands of words” generated by video sequences. Voluminous data notwithstanding, several key applications and industries are today turning to computer vision systems in search of efficient solutions to myriad problems. One such application that has tremendous potential is tracking.
The objectives of a tracking system must be
To track without regard to the environment, i.e. without imposing restrictions on the background/foreground
To track without requiring extensive a priori knowledge about background
To track target objects in the presence of similar objects in the vicinity
To track despite occlusions
To not require anything more than general purpose computing hardware i.e. a desktop personal computer.
The applications of such a tracking system include:
Autonomous Robotics – Robots in dynamic environments need to be able to interact with other moving objects. The ability to avoid collisions via motion estimation would allow robots to work freely in a changing environment
Automotive - In the quest to produce autonomous vehicles the first and foremost task is to sense moving objects around the vehicle and then to determine the speed and distance of the object. A visual tracking system can thus be of immense help to gauge the path in the case of such vehicles.
Surveillance - A camera could become a watchful eye over a shopping mall or airport. By integrating a tracking system and a facial recognition system, scans can be run electronically round the clock. Systems could possibly identify unwanted intruders or known felons from databases of thousands of images more quickly.
Education – With current emphasis on videotaping classroom lectures for broadcast, it creates resource challenges by way of requiring a cameraman to be present during all lectures. This can be avoided by using an automated system with an active camera. By tracking the teacher’s position in the class, the active camera could be used to keep him in focus wherever he moves.
Entertainment – In a virtual environment a computer could witness and interact with a user’s movements. Arcade games could accommodate the movement of their opponents by changing their screen or strategy to suit. One could dual against a machine without the need for external sensors attached to the body or motion with a hand to indicate a particular course of action to take.
Military experts have for long held that the best camouflaging systems fail when the target is in motion. While the human mind may overlook a stationary camouflaged object, the moment the object begins to move, it reveals itself. The brain processes information in such a way that even small changes in the field of view are immediately obvious. Thus, motion estimation and motion processing in general seem to be a vital part in the way the brain processes visual information. It was felt that we would do well to implement a tracking system on similar lines – using the properties of motion. Subsequent chapters deal with the tracking system in more detail.
To track without regard to the environment, i.e. without imposing restrictions on the background/foreground
To track without requiring extensive a priori knowledge about background
To track target objects in the presence of similar objects in the vicinity
To track despite occlusions
To not require anything more than general purpose computing hardware i.e. a desktop personal computer.
The applications of such a tracking system include:
Autonomous Robotics – Robots in dynamic environments need to be able to interact with other moving objects. The ability to avoid collisions via motion estimation would allow robots to work freely in a changing environment
Automotive - In the quest to produce autonomous vehicles the first and foremost task is to sense moving objects around the vehicle and then to determine the speed and distance of the object. A visual tracking system can thus be of immense help to gauge the path in the case of such vehicles.
Surveillance - A camera could become a watchful eye over a shopping mall or airport. By integrating a tracking system and a facial recognition system, scans can be run electronically round the clock. Systems could possibly identify unwanted intruders or known felons from databases of thousands of images more quickly.
Education – With current emphasis on videotaping classroom lectures for broadcast, it creates resource challenges by way of requiring a cameraman to be present during all lectures. This can be avoided by using an automated system with an active camera. By tracking the teacher’s position in the class, the active camera could be used to keep him in focus wherever he moves.
Entertainment – In a virtual environment a computer could witness and interact with a user’s movements. Arcade games could accommodate the movement of their opponents by changing their screen or strategy to suit. One could dual against a machine without the need for external sensors attached to the body or motion with a hand to indicate a particular course of action to take.
Military experts have for long held that the best camouflaging systems fail when the target is in motion. While the human mind may overlook a stationary camouflaged object, the moment the object begins to move, it reveals itself. The brain processes information in such a way that even small changes in the field of view are immediately obvious. Thus, motion estimation and motion processing in general seem to be a vital part in the way the brain processes visual information. It was felt that we would do well to implement a tracking system on similar lines – using the properties of motion. Subsequent chapters deal with the tracking system in more detail.
No comments:
Post a Comment