Analysis Of Pedestrian Attributes Based On Video Surveillance

In recent years,with the widespread public concern about social security issues,more and more video surveillance systems have been used in life.In the surveillance video,people are eager to obtain information about pedestrians,such as gender characteristics,age characteristics,height characteristics and so on.In traditional video surveillance,pedestrian attribute information is obtained by manual observation,but this method is only applicable to video with a small amount of data.For a large amount of surveillance video,using this method will consume a lot of manpower and material resources,and the accuracy is low,which is difficult to meet the actual application requirements.Therefore,it is necessary to carry out intelligent research on surveillance video,which is of great significance to the development of society.Pedestrian attributes are not completely defined.This dissertation mainly studies and analyzes the gender attributes and height attributes of pedestrians.For pedestrians in video surveillance,local feature detection is used for gender recognition,and height is estimated based on the imaging principle of the camera.The main research work is:Firstly,based on the hybrid Gaussian background modeling,this paper dynamically updates the learning rate of each pixel,and can also extract better foreground moving targets in the video with changing background.The extracted foreground moving targets are detected by the DPM algorithm,and the pedestrian area is separated from the complex area,which lays a foundation for accurate detection of subsequent attributes.Then,for the case where multiple frames appear in the same target in a video,the foreground pedestrians are tracked,and the problem that each frame is detected is avoided.In this paper,the tracking algorithms of Mean-shift,ASMS and KCF are tracked in different scenarios,and the KCF tracking effect is better than the other two.Therefore,the KCF tracking algorithm is used for pedestrian tracking.Next,a method for extracting the characteristics of pedestrian local regions is proposed to perform gender detection.The convolutional neural network and HOG are used to extract the local features of the pedestrian body,and then the two features are combined for gender detection,which is greatly improved compared with the traditional feature extraction accuracy.Finally,according to the imaging principle,the camera calibration is used to obtain the internal and external parameters,and then the main axis of the human body is obtained according to the feature vector corresponding to the center of the foreground region and its minimum eigenvalue.The intersection of the main axis and the foreground frame is used as the image of the pedestrian head vertex and the footing point.Pixel coordinates,the corresponding space coordinates are obtained according to the conversion relationship,and the height is finally solved.The method of analyzing pedestrian attributes proposed in this paper has a male gender recognition accuracy rate of 84.5%and a female gender recognition accuracy rate of 81.5%in the self-built database,and can effectively detect and identify pedestrians in the video,and estimate the height error.No more than 1.9%,which proves the effectiveness of the algorithm.