A pedestrian detection method for small target in complex scene

Abstract

The invention belongs to the technical field of pedestrian detection and relates to a small target pedestrian detection method under complex scenes. Firstly, a neural network is trained. The sample data is normalized in size; The shared feature extraction network is used to extract the shared feature, and the feature map is obtained. Classification feature extraction network is used to extract classification features from that feature map to obtain a classification feature map; the rough regression operation is carried out, the result of the rough regression is filtered by the confidence degree to obtain the candidate region; Classification of pedestrian target or not is carried out according to the classification feature map and the corresponding classification feature extracted from thecandidate area. Quadratic regression was performed on the candidate region, and the final regression positionis obtained by combining the position obtained by rough regression operation. SoftNMS is carried out by using the final regression position, and the duplicate box is removed to obtain the final target position and the corresponding classification result. The invention can effectively solvethe problems of high false detection rate and imprecise positioning of small target detection under complex scenes.

Description

technical field [0001] The invention belongs to the technical field of pedestrian detection, and in particular relates to a small-target pedestrian detection method in complex scenes. Background technique [0002] With the continuous progress of safe city construction and the continuous improvement of monitoring systems in various industries, a large amount of video image data is collected every second. In most surveillance videos, pedestrians constitute the main body of activities. How to make good use of these data to achieve accurate analysis of pedestrian trajectories and accurate extraction of pedestrian structural features has become an urgent problem in the field of security in the era of big data. [0003] Pedestrian detection technology is one of the most important core technologies to solve the above problems. Not only that, pedestrian detection also has very important applications in many other fields such as behavior analysis, regional prevention and control, p...

AI Technical Summary

Problems solved by technology

However, pedestrian targets, especially small pedestrian targets in complex backgrounds, have unique characteristics: single target category, high feature discrimination of large targets, and low feature discrimination between small targets and complex backgrounds

Images