Improved TLD Target Tracking Algorithm Based On Automatic Adjustment Of Surveyed Areas

Traditional TLD(tracking learning detection) is a semi supervised learning algorithm for target tracking architecture. In order to achieve the object can be tracked long-term, first of all, algorithm need to deal with complex background, distance scaling and light according to changes, partial occlusion problem and algorithm's time complexity to low to achieve the effect of real-time tracking.But algorithm the time of detection processing takes too long and the result of similar object tracking is not satisfactory. Another problem in TLD model is that it can't hand multiple-object tracking in every single video frame.Therefore in this paper, an efficient approach(TLD-DO) is proposed for object tracking in which the surveyed areas can automatic adjustment according to the object's velocity of movement. It also proposes a multiple-object tracking optimization algorithm based on TLD-DO in this paper. In order to accelerate the process speed of TLD algorithm without reducing tracking precision, a novel algorithm named DKF with optimal surveyed area which can reduce the detection range of TLD detector is constructed based on twice Kalman filtering operation for acceleration correction. At the same time, by the multiple-object optimization of tracker, learner and detector, multiple objects can be tracked in single tracking process of single video frame. Experimental results show that tracking effect of improved method is better than that of the original TLD algorithm under the circumstance of similar target disturbance. Furthermore, the detection speed has been improved between1.31xand3.19 xfor different video and scenes. At last, the multiple-object TLD-DO algorithm not only handles multiple-object tracking and inherits the characteristics of fast operation and strong anti-interference ability in TLD-DO, but also has a better identify processing capability that solve the problem in multiple similar objects tracking.