Detection and tracking are important components of many computer vision applications including automated surveillance. Object detection should overcome challenges such as changes in object appearances, illumination, and shadows. In our system, Gaussian mixture models are used for background subtraction to detect moving objects. Tracking is challenging because measurements from the object detection stage are not labelled and could originate from false targets. Our system uses multiple hypotheses tracking to solve the measurement origin problem. Practical long-term object tracking should have re-identification capabilities to deal with challenges arising from tracking failure and occlusions. To this end, each tracked object is assigned a one-class support vector machine (OCSVM), which learns the appearance model of that object. The OCSVM is trained online using HSV colour features. As a result, objects that were occluded or left the scene can be re-identified and their tracks extended. Standard, publicly available data sets are used to test the system.
Reference:
Keaikitse, M., Brink, W. and Govender, N. 2014. Long-term tracking of multiple interacting pedestrians using a single camera. In: PRASA, ROBMECH and AfLat International Joint Symposium, Cape Town, 27-28 November 2014
Keaikitse, M., Brink, W., & Govender, N. (2014). Long-term tracking of multiple interacting pedestrians using a single camera. PRASA. http://hdl.handle.net/10204/7907
Keaikitse, M, W Brink, and Natasha Govender. "Long-term tracking of multiple interacting pedestrians using a single camera." (2014): http://hdl.handle.net/10204/7907
Keaikitse M, Brink W, Govender N, Long-term tracking of multiple interacting pedestrians using a single camera; PRASA; 2014. http://hdl.handle.net/10204/7907 .