Modeling and Analysis of Video Image Target Detection (TD) and Tracking Based on Deep Learning

In recent years, deep learning technology and some related neural network models have shown strong learning ability in hierarchical feature learning, and have made great achievements in image classification, target recognition and scene understanding. Considering the strong hierarchical feature learning ability of deep network model, a correlation filter detection and tracking model based on dynamic spatial feature selection is proposed in this paper; In order to further improve the mathematical modeling method of the algorithm and improve the accuracy of the algorithm, a correlation filter detection and tracking model based on the fusion of spatial feature selection and channel feature selection is proposed; In order to optimize the mathematical modeling method of the algorithm and improve the real-time performance of the algorithm, an accelerated correlation filter detection and tracking model based on improved alternating direction multiplier method is proposed. The experimental results show that the depth feature of the algorithm proposed in this paper has excellent tracking accuracy and robustness, can be well applied to real-time tracking scenes, realizes model dimensionality reduction and improves the detection and tracking performance; In otb100 dataset, this algorithm achieves the optimal position prediction success rate under the 20 pixel error threshold and the tracking success rate under the 50% overlap threshold. The success rate curve srdcf is slightly higher than that of AUC algorithm; In templecolor128 dataset, although the algorithm proposed in this chapter only obtains the second overlapping accuracy and the third position accuracy, its tracking accuracy is very close to that of stable and srdcf algorithms; In terms of the real-time performance of the algorithm, the algorithm proposed in this paper can well meet the needs of real-time tracking.