Tracking in Multiple Cameras with Overlapping FOV
Multiple cameras are needed to cover large environments for monitoring activity. To track people successfully in multiple perspective imagery, one needs to establish correspondence between objects captured in multiple cameras. We present a system for tracking people in multiple uncalibrated cameras. The system is able to discover spatial relationships between the camera fields of view and use this information to correspond between different perspective views of the same person. We employ the novel approach of finding the limits of field of view (FOV) of a camera as visible in the other cameras. Using this information, when a person is seen in one camera, we are able to predict all the other cameras in which this person will be visible. Moreover, we apply the FOV constraint to disambiguate between possible candidates of correspondence. Tracking in each individual camera needs to be resolved before such an analysis can be applied. We perform tracking in a single camera using background subtraction, followed by region correspondence. This takes into account the velocities, sizes and distance of bounding boxes obtained through connected component labeling. We present results on sequences taken from the PETS 2001 dataset, which contain several persons and vehicles simultaneously. The proposed approach is very fast compared to camera calibration based approaches.
Saad Khan, Omar Javed, Zeeshan Rasheed, Mubarak Shah, Human Tracking in Multiple Cameras, 8th IEEE International Conference on Computer Vision, ICCV 2001, Vancouver, Canada, July 2001.
Omar Javed, Sohaib Khan, Zeeshan Rasheed, Mubarak Shah, Camera Handoff: Tracking in Multiple Uncalibrated Stationary Cameras, IEEE Workshop on Human Motion, HUMO-2000, Austin, TX, December 2000.
[Training Sequence, used to generate lines] [results] [symbolic results]
This sequence shows two persons in a room with three cameras. The FOV lines were established by one person walking in the room for about 40 seconds
[orig (small sample)] [symbolic results]
This sequence is taken from the PETS 2001 dataset. There are multiple persons and cars in this scene. A training sequence, which also comes with the dataset was used to establish the lines automatically, without the constraint of only one person walking in the environment. The given sequence is the testing sequence, in which correspondence was established correctly.
Tracking Results on Sequence 3:
One way of displaying results is to generate movies of tracked objects, with best view selection from the cameras in which the object is visible. To do this, correct correspondence must be established between different views of the same object, and hence is a good way to visualize the results. Below are some objects seen in Sequence 3.
[Person1] [Person5] [Person6] [Person7]