Convolutional neural network and support vector machine-based human body behavior recognition method

Abstract

The invention discloses a convolutional neural network and support vector machine-based human body behavior recognition method. The method comprises the following steps of: obtaining an RGB image comprising a person through a camera; carrying out feature extraction on the RGB image to obtain a group of convolutional feature map; inputting the convolutional feature map into a double-branch deep convolutional neural network to carry out processing, so as to obtain joint information and joint association information of the person in the RGB image, and generating human body skeleton sequence datathrough joint matching, wherein the joint association information indicates mutually associated information between adjacent joints; normalizing the obtained human body skeleton sequence data; and recognizing and classifying a human body skeleton sequence map through a multi-classification support vector machine. According to the method, human body behaviors are recognized through extracting and processing human body skeleton information, so that the method has favorable robustness and correctness; and real-time behavior recognition and analysis can be carried out on the basis of two-dimensional image data acquired by a conventional video monitoring system, so that the method has universality and practical significance for the application in the fields of intelligent security, production security and the like.

Description

technical field [0001] The invention relates to the technical fields of computer vision, machine learning and pattern recognition, in particular to a method for recognizing human behavior in RGB images based on a double-branch convolutional neural network and a multi-classification support vector machine. Background technique [0002] Human behavior recognition technology is an important branch and cutting-edge technology in the field of machine vision. It can be widely used in intelligent video surveillance, robot vision, human-computer interaction, game control, etc., and has a broad application market prospect. Various human behaviors can be characterized by the relative relationship of each bone joint point of the human body, so it is completely effective and feasible to identify the behavior of the human body by describing the points of human bone joint movement. At present, the behavior recognition based on human skeleton is carried out through the RGB-D image of the d...

AI Technical Summary

Problems solved by technology

At present, the behavior recognition based on human bones is carried out through the RGB-D image of the depth camera, but the depth camera contains a depth sensor, which is expensive and not suitable for ordinary monitoring scenarios, so it cannot be recognized for existing surveillance cameras.

At the same time, the current commonly used recognition method is to use a single person detector to detect and recognize the behavior of a single person, which has certain limitations. For example, if there are contacts between multiple people in the image, self-occlusion and object occlusion, detection and recognition are prone to occur. Error, and if there are too many people in the image, the detection time will be obviously too long

Therefore, the existing behavior recognition methods are difficult to promote and apply in the fields of video surveillance, robot vision, etc.

Images