The improved TLD algorithm that combines CAMShift and orientation prediction to realize face tracking in low illumination
Lin Zhang, Jin Hou, Mingju Chen, Hongwen Li · 2020
TLD(Tracking-Learning-Detection)algori thm acquires the target characteristics through continuous learning of the initial target. It can carry out long-term target Tracking. However, during the Tracking process, illumination, occlusion and target movement have a great impact on the Tracking effect. At the same time, the algorithm has a relatively large amount of calculation in the calculation process, which leads to the poor real-time performance of tracking. In order to reduce the influence of low illumination and occlusion on the algorithm, an improved algorithm integrating LBP algorithm into CAMshift algorithm was adopted to replace the original algorithm of tracking module in the algorithm. At the same time, in order to eliminate the defect of tracking loss caused by the target movement, Markov method is used to predict the direction and position of target movement , which can reduce the detection range of the algorithm and improve the real-time performance of the algorithm. The experiment shows that: compared with the traditional TLD algorithm, the improved algorithm has greatly improved tracking accuracy and faster tracking speed.