Optimization method of two-dimensional convolutional network for human body motion detection

Abstract

An optimization method of a two-dimensional convolutional network for human motion detection comprises the following steps for Yolov3: 1) firstly, generating a grey-scale map through data preprocessing of a to-be-detected picture or video frame; scaling is carried out according to the length-width ratio of the original picture, the pixel value of the scaled picture is pasted into the grey-scale picture, the grey-scale value of the part which is not pasted is kept unchanged, and the pixel value in the scaled picture is subjected to normalization processing; 2) the processed image data are sent to a Darknet-53 network, and features are extracted; a feature extraction layer of the Darknet-53 network performs down-sampling on an input picture, and the number of channels of a feature map on each scale is twice that of a feature map on the previous scale; extracting features by each feature extraction layer, and performing Conv2D operation to obtain a feature map; and for the channels of the last three scales, features of the corresponding channels are predicted respectively, and then the prediction results are fused with the prediction results of the original channels and then output. According to the invention, the detection precision of the two-dimensional network model is improved.

Description

technical field [0001] Aiming at the field of motion detection, the present invention designs a human body motion detection method based on a two-dimensional convolutional neural network to obtain the category of target object behavior. [0002] technical background [0003] With the advancement of science and technology and the improvement of computer performance, artificial intelligence technology has been widely used. As an important technology in human-computer interaction and intelligent video surveillance, human behavior detection has always been a research hotspot in the field of computer vision. In the actual detection process, there are problems such as complex background, occlusion, various actions, and high acquaintance between actions, which makes the detection task more difficult. The behavior detection algorithm based on the convolutional neural network has the characteristics of strong feature extraction ability and high recognition accuracy in complex scenes,...

AI Technical Summary

Problems solved by technology

[0004] The invention provides a method for constructing a human body action detection model based on a two-dimensional convolutional neural network to solve the problem of low detection accuracy of the existing two-dimensional network model

Images