Research On Target Tracking Algorithm Based On Tracking-Learning-Detection

The video target tracking technology in computer vision field is a relatively cutting-edge research field,and can be widely applied in road safety,artificial intelligence,military defense,unmanned driving,human-computer interaction and other fields.Therefore,the study on moving target tracking technology is of considerable economic value and a wide range of application scenarios.Since the TLD(Tracking-Learning-Detection)target tracking algorithm was proposed by Czech scientists,it has attracted the attention and research of a large number of researchers.The TLD target tracking algorithm is an algorithm that can carry out long-term tracking of the target.However,it may fail to track the target when the target is occluded,deformed,or when there is illumination variation.In response to the above-mentioned problems,the following improvement measures were put forward.First,combining the Kalman filter with the TLD algorithm.The Kalman filter can predict the target motion position of the target in the next frame based on the information of the previous frame of the moving target,and update the TLD algorithm with the predicted target position as the observation value,so as to ensure the accuracy of the tracking process.Second,combining the CamShift algorithm with the TLD algorithm.Because the CamShift algorithm has good anti-interference ability for moving targets with varying scales,and is capable of calculating the target model vector,that is,the target's moving direction,so it can be used to update the model for the TLD algorithm.Third,using color features and HOG(Histogram of Oriented Gradient)features as features for TLD algorithm tracking.By calculating the HOG features in the direction of the edge of the image,the contour features of the image can be accurately grasped,thus greatly reducing the influence of illumination on the tracking process.In addition,this paper extended TLD single-target tracking to multi-target tracking,which was then transplanted to the embedded Raspberry Pi 3B+ platform to meet the miniaturization requirements of the tracking system.The analysis of experiments and comparison between the traditional TLD algorithm and the improved TLD algorithm shows that the improved TLD algorithm in this paper is capable of accurately tracking the target under similar target interference,occlusion,deformation and illumination variation,with improved accuracy and robustness of the algorithm.