Abstract: Traditional tracking-learning-detection (TLD) object tracking algorithm takes long time for detection because the area to be detected is too large， and the algorithm is not satisfactory for tracking similar objects， and it is only suitable for single object tracking tasks. Therefore， an efficient TLD-detector optimization for multiple objects (TLD-DOMO) approach is proposed for tracking multiple objects， which is built on a novel algorithm named DKF (double Kalman filter). Detection areas are adjusted adaptively by the prediction method accelerated by double Kalman filtering operation. The multiple-object detectors of TLD-DOMO algorithm can predict potential motion range of each object to optimize the scale and position of detection areas adaptively. Thus， the accuracy and efficiency of multiple-object tracking will be improved. Moreover， the proposed method also reduces the interference among tracked objects effectively for supporting similar objects tracking. Experimental results show that the detection efficiency is improved in all test videos， and the speedup ratios are between 155% and 294%. The effect of detection and recognition of multiple objects surpasses the original TLD approach in tracking multiple similar objects.

Key words: object tracking, TLD algorithm, multiple-object, detection area, tracking speed