Target Tracking Based On ResNet Motion Scene Classification

As one of the basic tasks of computer vision,visual target tracking is widely used in intelligent transportation,security monitoring,trajectory prediction,pedestrian recognition and so on.There are many difficulties in the actual scene of target tracking,such as occlusion,low resolution,small tracking target and long tracking time,which will affect the tracking performance.Therefore,it is of great significance to make target tracking faster and more accurate.In order to improve the performance of target tracking,this paper improves the target tracking algorithm on the feature extraction of tracking network and the integrated selection of tracking model.The main innovations of this paper are as follows.(1)In the process of feature extraction in target tracking network,the extracted features have many channels,and each channel also has many eigenvalues,and each feature has different influence on the tracking results,but the importance of these features is not distinguished in the tracking algorithm.To solve this problem,this paper proposes an improved Siam FC target tracking algorithm based on CBAM attention mechanism.In the improved algorithm,these eigenvalues will be weighted,so that the important features get a larger weight value,and the unimportant features get a smaller weight value,so that the size of the weight value is directly proportional to the importance of the features,so as to improve the tracking performance.(2)In target tracking,the moving scene of the target is complex and diverse.The ability of target feature extraction in the same scene varies greatly when tracking networks with different structures,resulting in obvious differences in the performance of different tracking algorithms when tracking targets in the same scene.The tracking effect of the network with poor feature extraction ability is not ideal in the actual scene.To solve this problem,this paper proposes a target tracking algorithm based on target motion scene classification,which can show excellent tracking ability in a variety of scenes.The algorithm first uses the residual network of transfer learning to classify the target motion scene,and then selects the appropriate network model to track the target according to the classification results.The first mock exam is conducted by using UAV123 dataset.The experimental results show that the method has higher tracking success rate and accuracy than single model.Compared with other trackers on OTB100 data set,the tracking effect is better than other trackers.