Pedestrian detection and tracking based on TLD framework

董永坤,王春香,薛林继,杨明
2013-01-01
Abstract:A computer vision based method was proposed to detect and track pedestrians. It worked on TLD (tracking-learning-detection) framework which was a target detection and tracking algorithm that runs the detection and tracking simultaneously and then integrated the results to update the detection and tracking modules. HOG (histograms of oriented gradients) operator and an incremental classifier based on NCC (normalized cross correlation) matching were proposed to improve the detector for the reason that TLD was prone to an error or failure when it was applied to track a pedestrian. Firstly HOG descriptor was calculated to match with the templates stored in classifier based on NCC, then the results were integrated with tracking module, and final results will be used to update the detecting module. According to the experimental results, this method has good robustness and efficiency.
What problem does this paper attempt to address?