Abstract:

Aiming at the problems of backward pedestrian statistics, non-real-time and backward statistical data, intelligent video surveillance and image recognition were proposed for real-time pedestrian traffic statistics. The system calculated the pedestrian target characteristics according to the integral channel idea, and used the Adaboost algorithm to train the classifier to locate and identify the pedestrian target in the image frame. Based on the identified targets, the CPU multi-task model was used to improve the kernel correlation filtering algorithm to track the target in real time and obtain the pedestrian traffic. The test results showed that the system can recognize, track and count pedestrian targets in real time. The average recognition rate was 93%, and the improved multi-task model increased the tracking rate by about 20%.

Key words: pedestrian flow statistics, real-time monitoring, integral channel features, kernel correlation filtering, multi-task tracking, Adaboost algorithm