Night-vision video gait recognition method based on angle radial transformation and centroid

Abstract

The invention provides a night-vision video gait recognition method based on angle radial transformation and centroid. The method includes the following steps that: infrared video images are acquired; human body gait contours are extracted from the acquired infrared video images, and the extracted human body gait contour sequence is pre-processed; the gait energy graph of a human body is built according to the pre-processed human body gait contour sequence; the feature vectors of the gait energy graph is extracted, the angle radial transformation coefficient and centroid vector of the gait energy graph are adopted as the feature vectors; and the dimensionalities of the feature vectors are reduced through the principal component analysis method; and recognition matching is performed on the dimensionality-reduced feature vectors and a training sample set, so that the gait recognition of the infrared night vision images is realized. With the method provided by the invention adopted, the effective information of the gait energy graph of the night vision video images can be extracted effectively and effectively, the efficiency of the gait recognition can be improved, and the correct rate of the gait recognition can be improved.

Description

technical field [0001] The invention relates to a night vision video gait recognition method based on angle radial transformation and centroid. The method belongs to the technical field of infrared night vision video image processing, and pedestrian monitoring at night can be realized through the method. Background technique [0002] In the past few years, infrared imaging technology has developed rapidly, and the application of infrared video images has become more and more extensive. For example, video surveillance at night protects the lives and properties of residents by automatically identifying family members and intruders. However, the infrared camera obtains a grayscale image, which has the characteristics of less texture details and low signal-to-noise ratio. Since the gait recognition can be based on the overall characteristics of the person, the video quality requirements are not high. Therefore, gait recognition is very active in infrared video recognition rese...

AI Technical Summary

Problems solved by technology

Model-based methods usually require obtaining a clear gait sequence and building a model through complex calculations to achieve good recognition performance, and the recognition rate drops significantly when the body is self-occluded

The second category is based on non-model methods. For example, Han et al. proposed gait energy map (GEI) to reflect the change of silhouette shape. This algorithm is more robust to noise, but it is not enough to reflect the dynamic changes between consecutive frames. Part of the gait information is lost; Wang Liang et al. proposed a contour-based unwrapping gait recognition algorithm, which uses the line between the contour composition points and the center of mass of the human body to express the gait features. This method depends on the overall human body The change of contour shape over time has a certain impact on the recognition rate when clothing or arm posture changes

The gait recognition algorithm based on the center of mass proposed by Chen Xin et al. takes the fluctuation of the center of mass as the gait feature, and uses the frequency spectrum analysis of the center of mass trajectory to identify individuals. It has achieved a high recognition rate and a small calculation cost, but the center of mass as the only feature is vulnerable to Noise interference, and the similarity of centroid fluctuations are prone to occur when the number of samples increases, resulting in an increase in the difficulty of recognition

Images