Pedestrian target detection and recognition method based on monocular vision and deep learning

Abstract

The invention belongs to the technical field of pedestrian target detection, and discloses a pedestrian target detection and recognition method based on monocular vision and deep learning. The methodincludes: establishing a small sample pedestrian data set, and collecting road pedestrian images in a real scene; performing pedestrian detection based on the whole image candidate and a single regression target detection algorithm based on depth features; finely adjusting weight parameters of a higher layer of the network on the VOC data set and the small sample pedestrian data set through secondary transfer learning; based on multi-scale pyramid image feature extraction with consistent phases, extracting contour features of pedestrian images, and obtaining a multi-scale pyramid feature map;and adopting a balanced focus loss function to replace a cross entropy loss function to measure the classification accuracy of the target. According to the invention, the CNN is used to obtain depth features, a deformable component model is trained, and the detection precision is effectively improved; transfer learning is introduced and can be found by analyzing a hidden layer in the AlexNet model, and the accuracy of pedestrian target detection and recognition is improved.

Description

technical field [0001] The invention belongs to the technical field of pedestrian target detection, in particular to a pedestrian target detection and recognition method based on monocular vision and deep learning. Background technique [0002] At present, the existing technologies commonly used in the industry are as follows: pedestrian detection has important applications in the directions of smart cars, robots, and video surveillance; Detection remains a challenging topic in the field of computer vision. At present, pedestrian detection based on computer vision is mostly based on feature extraction and machine learning methods; in terms of feature extraction, features such as contours, textures, frequency domain information, and color regions are often used to describe the difference between pedestrians and backgrounds. For example, features such as HOG, EOH, Edgelet, and Shapelet describe the outline features of pedestrians, LBP features describe the texture features of...

AI Technical Summary

Problems solved by technology

[0003] To sum up, the problems existing in the existing technology are: the traffic situation under the real road camera is complex, the flow of people is dense, and the distribution of the flow of people is unbalanced, resulting in an imbalance between the vehicle sample classes, which is very different from the flow of people in the public data set, resulting in unbalanced model training. Learn the characteristics of different types of people flow, and the detection effect of other types of people flow is poor

Images