Multi-feature fusion bird's-eye view pedestrian detection method based on aggregated channel features and gray level co-occurrence matrix

Abstract

The present invention relates to a multi-feature fusion overlooking pedestrian detection method based on aggregated channel features and gray-level co-occurrence matrix, comprising: extracting ACF features of multiple aggregated channels in a sample training set, obtaining aggregated channel feature vectors and gray-scale co-occurrence matrix feature vectors, Send the two into the soft cascade Adaboost classifier for training to obtain classifier 1 and classifier 2; read the image to be tested, extract its ACF features, and obtain the aggregate channel feature vector; send the aggregate channel feature vector to classifier 1 for classification , to obtain the candidate coordinates and the target window; to obtain the eigenvector of the gray level co-occurrence matrix, and send it to the second classifier to eliminate the background interference, and obtain the output result of the final target. The invention fuses the color, gradient direction histogram, gradient and texture features, filters out the background similar to the human head, effectively reduces the missed detection and false detection rate of the classifier, improves the detection performance of overlooking pedestrians when there are many interference backgrounds, and is stable and reliable And efficient, has strong practical application value.

Description

technical field [0001] The invention belongs to the technical field of computer vision pedestrian detection, in particular to a multi-feature fusion bird's-eye view pedestrian detection method based on aggregated channel features and a gray-level co-occurrence matrix. Background technique [0002] At present, the overlooking pedestrian detection technology in intelligent monitoring scenarios is widely used in people flow statistics and intelligent analysis of pedestrians, so as to achieve effective supervision of crowded places. Look-down-based pedestrian detection has made rapid progress in recent years. However, the pedestrian's hairstyle, hair color and complex surrounding background increase the difficulty of detection, resulting in unsatisfactory detection results, which need further improvement. There are several categories of looking down pedestrian detection methods. The first type is a method based on shape analysis. The main idea is to screen image samples one by...

AI Technical Summary

Problems solved by technology

Although the current overhead pedestrian detection method based on multi-feature extraction can reduce the false detection of overhead pedestrian detection in the background of interference, the degree of improvement in algorithm performance is limited, and the combination of multiple single features will also increase the computational complexity.

Images