An Improved TLD Tracking Method Using Compressive Sensing

Qiang Li, Xueshi Ge, Geng Wang
2016 Proceedings of the 2016 International Conference on Artificial Intelligence: Technologies and Applications   unpublished
Visual Tracking, as an important subject in computer vision, has been widely used in surveillance, space exploration, and human-computer interaction etc. Both tracking-learningdetection (TLD) [1] and compressive tracking (CT) [2] are successful algorithms among those proposed recently. However, TLD suffers from low efficiency and CT overlooks scale change during tracking. In this paper, we propose an improved TLD tracking algorithm by using compressive sensing. The improvements include
more » ... the detection method in TLD with CT, employing Kalman filter in detector to estimate the tracking region for improving the detection speed. Besides, adaptive search radius is employed to deal with object disappearance and shielding issue. Lastly, the tracking results of TLD and CT are integrated to estimate the target status and update the classifier. The experiments show that, compared to the original algorithms, the improved algorithm combines the advantages of two algorithms, conducing to accurate tracking precision, faster tracking speed and handling the object extent change.
doi:10.2991/icaita-16.2016.64 fatcat:cjlhnuplh5c5vgtm6xcz5po3vu