Pedestrian Tracking Algorithm for Siamese Networks Based on Improved Attention Mechanism and MASK R-CNN Detection

Pedestrian tracking is an important application direction in the field of visual object tracking. Due to the high flexibility of pedestrians, inaccurate information about the initial position and size of pedestrians provided to the object tracking algorithm in complex situations can lead to degradation of the performance of the object tracking algorithm as well as tracking failures. In order to solve the problem of tracking performance degradation and tracking failure of pedestrian tracking algorithms due to the inaccurate initial position and size information of pedestrians provided in complex situations as well as the decrease of network attention, this paper proposes a Siamese network pedestrian tracking algorithm based on an improved attention mechanism and MASK R-CNN detection. In this, this paper has done the following four research works: first, the MASK R-CNN detection model is used in the algorithm to detect the pedestrians; second, the backbone feature extraction network and the attention mechanism are improved in the algorithm; third, the method of ordering the number of pedestrians is designed in the detection model; and fourth, the results of the two convolutional inter-correlations are fused in the algorithm. This algorithm is divided into 2 parts, the first part is the detection model, which is trained on PenFudanPed dataset, and the second part is the tracking model, which is trained on GOT-10K dataset, and finally, the pedestrian object in the object tracking benchmark dataset OTB100 is selected to test the algorithm for evaluation, and the test results show that the algorithm performs excellently in comparison with the other excellent algorithms.