Pedestrians Segmentation And Its Application In Tatic Image

Pedestrian segmentation has important value in video surveillance. It is one of the hot spot of image research in recent years. Pedestrian segmentation of static image can be applied to many applications, such as image compression, image editing, image searching, pedestrian gesture recognition and criminal forensics. In surveillance scenes, clothes are used as an implicit identification clue to look for the pedestrian target on an event focused image. Clothes color recognition has been widely used in many aspects. With smart clothes and apparel from referrals, clothes color recognition can be used to improve pedestrian recognition, video surveillance, computer iconography, and content-based image retrieval research. This thesis analyzes the existing algorithms on corresponding areas, focusing on the pedestrian segmentation in static image and identifying pedestrian clothes color.The main contributions and novelties in this thesis are as follows:(1) The currently existing pedestrian segmentation methods and the relevant methods of pedestrian clothes color identification were reviewed. The research background, significance and research status of pedestrian segmentation and clothes color recognition are introduced, and the main work and content of this thesis is briefly described.(2) The Graph Cut algorithm, the Level Set algorithm and some other traditional image segmentation methods and some feature classification methods were mainly introduced. This thesis analyzes these algorithms, then the methods of this thesis combined with the bit and disadvantages of these algorithms were put forward.(3) In view of the Graph Cut image segmentation can’t get the results meet the requirements of the semantic faults, a improved Graph Cut algorithm combining shape priors and discriminatively learned appearance model was proposed in this thesis to segment pedestrians in static images. In this approach, a large number of real pedestrian silhouettes were used to encode the apriority shape of pedestrians, and a hierarchical model of pedestrian template was build to reduce the matching time, which will hopefully bias the segmentation results to be humanlike. A discriminative appearance model of the pedestrian was also proposed in this thesis to distinguish persons from the background better than original method. Finally, it shown the segmentation results of the method compared with the results of traditional Graph Cut segmentation algorithm. Experimental results verify the improved performance of this approach.(4) On the basis of pedestrian segmentation in the third chapter, the human clothes color recognition was studied. Gaussian model and color histogram were extracted to the color feature, and the minimum error rate Bayesian decision, K-nearest neighbor (KNN) algorithm and support vector machine (SVM) were used to classify the clothes color features, so as to identify the pedestrian clothes color. Finally, we analyzed the results of the methods. Experimental results show that the methods can identify the color of clothes for pedestrians more accurately. Our method can meet the application requirements.