Pedestrian Tracking Algorithm for Dense Crowd based on Deep Learning

The development of Dense Crowd Visual Tracking algorithm based on Deep Learning (DTDL) is introduced. The main research contents of this paper are as follows: (l) Dense crowd Based on Single Shot Multi-Box Detector (D-SSD). The D-SSD detection algorithm was proposed to reconstruct the pedestrian data set, mark the head and shoulder of pedestrians in the crowd, and modify the size of the candidate box close to the proportion of head and shoulder of pedestrians, so as to obtain a new data set for training. (2) visual tracking framework (Kalman Based on Kernel Correlation Filters, K-KCF): KCF tracking framework is more popular in recent years, KCF used in visual tracking under the common scene both in tracking performance and tracking speed performance is good, but there are severe target block KCF will tracking failure, in the dense population under the background of such inevitable condition. The addition of Kalman filter can largely solve the tracking failure caused by the occlusion of pedestrians. In this paper, Kalman filter is proposed as an auxiliary K-KCF visual tracking framework.