Subway section passenger flow statistics and pedestrian retrograde motion detection method and system

Abstract

The invention discloses a subway section passenger flow statistics and pedestrian retrograde motion detection method based on a multi-target tracking algorithm. According to the invention, an online multi-target tracking algorithm is applied to a subway monitoring scene, a subway monitoring video is used to make a sample set for training a detector, and a large-scale pedestrian re-identification data set is used to train an appearance feature extraction network. The performance of a tracking algorithm is improved by integrating motion information and appearance characteristics, each pedestriantarget in a scene is tracked through the algorithm, then passenger flow information is counted through logical judgment between the tracking target and a set trip line, and meanwhile, the retrogradebehavior of a traveler can also be detected.

Description

Technical field [0001] The invention relates to the technical field of pedestrian flow and abnormal behavior detection, in particular to a method and system for subway section passenger flow statistics and pedestrian retrograde detection based on a multi-target tracking algorithm. Background technique [0002] With the continuous advancement of my country's urbanization and continuous investment in the field of infrastructure construction, large cities in the first and second tiers have planned relatively complete rail transit networks. For the city, the subway not only greatly facilitates the daily commuting of the people, but also improves the comprehensive management level of the city. At the same time, as a typical crowded place, the subway also has many new safety issues, and its public safety governance also puts forward higher requirements for managers. Using video surveillance to control passenger flow and warn of dangerous events can not only ensure passenger safety, bu...

AI Technical Summary

Problems solved by technology

This method cannot be applied in real time because it needs the data of the front and rear frames of the video, and the motion vector direction of the continuous frames is nonlinear, so the practicability is not good.

However, by calculating the relative change in the Y coordinate of the pedestrian's centroid position and the relative change in the width-to-height ratio of the pedestrian detection frame in the three frames before and after, it is judged whether the pedestrian has fallen abnormally, which will cause an error due to the instantaneous change of the detection frame due to a large number of occlusions. Judgment, poor practicality

Images