Crowded scene pedestrian target detection and counting method based on density estimation.

Abstract

The invention provides a crowded scene pedestrian target detection and counting method based on density estimation. The method comprises the following steps: firstly, converting an input marked image into a continuous density map by utilizing Dirac function and Gaussian kernel function convolution; then, constructing a VDNet crowd counting network containing an expanded convolutional layer, and inputting an original image and a continuous density map to train the network; and obtaining a generation density map of the input crowded scene image by using the trained network, and obtaining the predicted number of people of the image. According to the method, expansion convolution and density estimation are combined, more effective features can be extracted, the method has higher detection accuracy and robustness, and the method can be effectively applied to pedestrian target detection and counting in a crowded scene under a complex condition.

Description

technical field [0001] The invention belongs to the technical fields of computer vision and graphics processing, and in particular relates to a method for detecting and counting pedestrian targets in crowded scenes based on density estimation. Background technique [0002] Pedestrian target detection and counting in crowded scenes plays an important role in ensuring the safety of public places and preventing stampede accidents due to overcrowding. However, due to the fact that in a crowded scene, there is a problem that a single person in the picture taken from a long distance by the camera has few pixels and cannot be recognized. At the same time, because the crowds are too dense and mutual occlusion is serious, identification is more difficult, and it is more difficult to generate high-quality density maps, and the quality of density maps will directly affect the accuracy of prediction results. Therefore, there are relatively few studies on pedestrian target detection and...

AI Technical Summary

Problems solved by technology

The present invention combines dilated convolution and density estimation to explore how to increase the receptive field of the convolution kernel while keeping the parameter amount and calculation amount unchanged, while ensuring that the size of the output feature map does not change, and retains more Multi-features, with high accuracy and strong robustness, can effectively solve the problem of pedestrian target detection and counting in crowded scenes under complex conditions, with strong practicability

Images