Railway Foreign Object Tracking Based on Correlation Filtering of Optimized Regularization Model

Abstract

Aiming at problems such as the untrustworthy association between spatial regularization weight and intrusive foreign object in complex railway scenes, as well as the degradation of correlation filter model, fully excavate the expressive ability of deep space features, and a foreign object tracking algorithm based on correlation filtering with depth space and time perception regularization is put forward. Firstly, select the fifth-level convolution feature of the Visual Geometry Group (VGG) network to extract the spatial area information of the foreign object, which is used to solve the regularization guide weight. Secondly, a regularization term based on depth space is added to the objective function, whose aim is to establish a more reliable association between the spatial regularization weight and the invading foreign object. Thirdly, the time perception term is added to establish the connection between the filters in time. Finally, based on the depth space, a simple and effective model update strategy is proposed. On the public OBT datasets and complex railway scenes, the tracking results of the algorithm in this paper and the existing multiple algorithms are compared and analyzed. The results show that in complex railway scenes, the algorithm in this paper is superior to other algorithms in distance accuracy and success rate. The tracking speed is 23.1FPS, which basically meets the real-time requirements. Therefore, the correlation filtering algorithm of the improved regularization model is of great significance to railway safety.

Keywords

• railway foreign object tracking
• correlation filtering
• depth space
• time perception
• spatial regularization
• model update