Pedestrian Recognition In The Template Selection And Feature Extraction

This article is based on the application of gait recognition. It mainly concerns on the gait templates and feature extraction from gait sequences. After years of re-search of many scholars, gait recognition has been divided into two categories:model-based and model-free. The model-based methods concerns on building structures of human body while model-free methods concerns on how to extract features from over-all views. Compared with model-based methods, model-free methods are more simple and compute-convenient. The proposition of Gait Energy Image (GEI) brings great in-fluence to model-free methods. GEI averages all the frames of one gait sequence into one single image, this action greatly decreases the load of computation and reduces a lot of noise.GEI performs outstandingly in computation by averaging all the frames with the same weight, but it ignores the temporal information absolutely. The Chrono-Gait Im-age (CGI) preserves the temporal information by mapping gait frames into RGB spaces by a specific color mapping function. The newly produced energy image with color contains both spatial and temporal information. The experimental results prove its ad-vantage by enhancing the accuracy rate of gait recognition.This article firstly concerns on some gait feature extraction technologies. Assuming that the high frequency information is critical in discriminating different people, we use Radon Transformation and the Differential Radon Transformation (DRT) to extract high frequency information and enhance the accuracy of gait recognition.The Histogram of Orientation Gradients (HOG) is a commonly used descriptor in image processing. In this article we propose a new Multiple HOG Templates feature for Gait Recognition by combining the advantages of HOG and GEI and CGI. It preserves both spatial and temporal information and performs better than GEI and CGI templates themselves from bundles of experiments.