New Method Of Human Body Gesture Recognition

Image-based human pose recognition technology is a new research field based on computer vision,which provides a technical basis for the development of intelligent monitoring and human-computer interaction.Through the processing of the image,the useless information can be ignored,and only the human body information can be extracted.According to some characteristics of the human body target,the human pose can be analyzed and recognized,and the human pose can be judged automatically.The computer has the function of intelligently understanding the human pose in the image.This paper mainly studies the pose detection and recognition technology of human body in fixed scene.Firstly,a target extraction technology is improved.Based on discussion and comparation of several main target detection techniques,a hierarchical and segmented.background model estimation method is proposed,which can extract human targets well.Then the human target feature will be extracted,the first step is that the edge detection method is used to extract human edge features.The second step is that the regional features are extracted from the results of the first segmentation,the coordinate system is established with the center of the extracted human body region as the origin,and the tangent value of the angle between the centroids of the four quadrants of the image and the centroids of the image and the x axis is established.The ratio of width to height,eccentricity,square of outline length and its area ratio are used as feature vectors to reflect human pose characteristics and train classifiers of pose recognition.In the last part,according to the feature vector extracted from 3000 sample images,BP neural network is used to train BP neural network according to standing,bending and falling posture,and the BP network model is established.BP neural network model is used to identify human pose.