Dual-learning-based method for counting pedestrians at subway station scene

Abstract

The invention discloses a dual-learning-based method for counting pedestrians at a subway station scene. The method is characterized by comprising the steps of (1) based on the subway video foreground extraction method of a Gaussian hybrid model, extracting the active computing area of a video pedestrian group for each frame of the video; (2) based on the dual-area division method for the subway monitoring video, dividing each frame of the video into a far-view area and a near-view area; (3) based on the far-view pedestrian estimation method of a super-strong learning machine (ELM), extracting the global scene characteristics of the far-view area, and establishing a pedestrian number estimation model based on the ELM and a Gaussian process regression model; (4) conducting the pedestrian number estimating method for pedestrians in the near-view area based on aggregated channel characteristics (ACF). According to the technical scheme of the invention, the video data acquired by each single monitoring camera in a subway station is adopted as the input, and the number of pedestrians obtained through counting pedestrians in each frame of a video data stream is adopted as the output. The counting result of the method is good in accuracy and real-time performance.

Description

technical field [0001] The invention relates to digital image processing counting, pattern recognition and visual computing, in particular to a method for counting pedestrians in a subway station scene based on dual-area learning. Background technique [0002] The continuous population growth and the intensification of the urbanization process have led to an increase in the frequency of sudden public pedestrian safety incidents such as pedestrian hedging and pedestrian stampedes. For such problems, the traditional solution is to install a large number of monitoring equipment in public places, and at the same time send additional manpower for uninterrupted attention. However, such approaches usually have the following disadvantages: [0003] 1) With the continuous increase of monitoring equipment, the information to be processed will also increase, which greatly increases the labor cost. [0004] 2) The amount of data is too large to rely on traditional manpower to analyze ...

AI Technical Summary

Problems solved by technology

[0003] 1) With the continuous increase of monitoring equipment, the information to be processed will also increase, which greatly increases the labor cost

[0004] 2) The amount of data is too large to rely on traditional manpower to achieve timely analysis, so that it is impossible to make timely responses to the situation on site

However, in the subway station scene, there are serious pedestrian occlusions and perspective transformations in the relevant surveillance camera images, so the above two methods cannot effectively obtain the number of pedestrians in the subway station scene.

Images