|
The moving target object classification is one of important research content in Computer image processing, Video-based motion analysis and Intelligent video surveillance processing.With the rapid economic development and the promotion of national building policy on urbanization, lead to social public security management, urban traffic management under severely test, the traditional security systems are facing unprecedented challenges, then the high-tech safety precaution equipments are being concerned about, such as video surveillance. The development of digital and network making video surveillance, those with the intuitive, accurate, timely and information content-rich features, is no longer merely to serve national defense, government offices, banks, large warehouses and some other with larger hidden factor and more sensitive sites, but also been widely used in directly contact with the people’s lives environment, such as urban infrastructure, urban road traffic management, public order management etc. And video surveillance can no longer depend only on the monitor’s day duty and on-site operations, in order to change the passive management situation, the technological development of intelligent video surveillance and analysis become more and more attention by the academia and the application field.Object classification is finding a certain type of target object’s essential characteristics from the current image information, so that they can only or approximately characterize a certain type object,and then able to distinguish such type object from the others. It including the two core issues:the object feature extraction and classifier construction. Objects classification involved in the study field of computer vision, image processing, pattern recognition and other research fields. Based on urban road video surveillance, this work is studying on the classification of common moving objects such as pedestrian, vehicles, the vehicle objects including cars, vans, buses and trucks. And on that basis, this paper carried out tentative classification study on car class.This article contents:1. The target image preprocessing, then through the comparative study of existing classification characteristics and methods, decide the paper’s classification method. On this basis, the paper does further study of the object features, then defined one new classification feature for cars.2.By analyzing and summarizing several existing classifier algorithm, determined the appropriate classifier used in this paper, that is, the statistical learning theory and structural risk minimization theory-based support vector machine method.3.To achieve an accurate classification, this paper presents a new object classification strategy, namely, a binary tree and Decision Directed Acyclic Graph combined hierarchical classification strategy. Use this in conjunction with the multi-feature classification, it can comply with the cross problem between the different object classes on certain degree.4.Finally, according to the method and strategies proposed in this paper to experiment, and listing the method, procedures and results of each experiment. Through experimental studies on common objects of the road video surveillance show that the method used has a certain effect, to achieve the common object classification target.
|