Research On Gait Recognition Method In Video Surveillance

With the development of science and technology, the awareness of security has been enhanced, which lead that a lot of monitoring infrastructures are widely mounted in public. It’s getting more and more important to identify people by the video information of the acquisitions from monitors. Gait recognition, relying on people’s walking posture to make the identifications, is of non-invasion, non-contact, easy concealment. All of its characteristics are very suitable for intelligent monitoring. Therefore, gait recognition gets widely concerned and researched.Gait recognition is studied by the video of people’s walking posture, whose hardcore is mainly about extracting suitable gait feature which takes advantage of classifier to express people’s walking characters. Gait recognition involves many fields, including digital image process, computer vision, pattern recognition, etc.Now, the extracting of feature of gait recognition is the main task in the whole study. This paper proposes a kind of manner by fusing the characteristics of partial contour and GEI, and mainly details the extraction and fusion. The research and innovation of this paper is put as follows:(1)method based on feature extraction of gait contour is proposed. First, background model is established with the vibe algorithm, uses the background subtraction method to get the foreground image, using morphology and connectivity analysis and processing, according to the ratio of width to height of the key frame extraction measured gait cycle; then counterclockwise, the contour of the human body to start the calculation of all contour points to the centroid distance, namely two-dimensional contour shape conversion for the one-dimensional signal of distance, the gait contour feature.(2)Extraction of gait skeleton features. Using an evolutionary skeleton extraction method to extract the skeleton features, then the extracted skeleton is refined, and then a new corrosion and reduction method is used to obtain the main skeleton of the gait, and then the skeleton of the human skeleton is obtained. Because the lateral shift of the skeleton point relative to the center of gravity is periodic, the Fourier series of the skeleton points can be obtained by mapping the Fourier transform to the frequency domain. The motion and shape of a sequence of steps can be represented by the low frequency Fourier series and have enough recognition ability.(3)The fusion of two kinds of features is carried out by using weighted addition. In the process of integration, the two features are different, and the reliability of the matching value is different in the process of fusion. In order to give full play to the advantages of each feature, the introduction of weights, so that the fusion results to achieve higher recognition rate, and finally the nearest neighbor fuzzy classifier for classification and recognition.Experiments show that the method is effective and useful in the application of gait recognition.