Research Of Vehicle Recognition Systembased On Dm8168 Dual-core DSP

Vehicle detection and recognition, the important parts of intelligent transportation system, are designed to acquire and analyze the information of the moving targets quickly and accurately, and make a critical difference in a road surveillance system. The research content is originated from Research of Network Image Acquisition and Analysis System Based on DM81 x Series Chip in Jiangsu Yitong High-tech Company Limited. In this paper, locating and recognizing the logo and the front view of a vehicle from high-definition video sequence images captured by traffic monitoring are presented as the following aspects:To begin with, having found the rough area of logo through the relative position between license plate and logo, we study the Sobel edge information and the morphological characteristics to determine the accurate location of logo. Besides, a horizontal gradient projection method is put forward to determine the desired block of vehicle face. The experiments about five hundred vehicle images show that the detection rates of the logo and the face are about eighty percent and eighty-five percent.Furthermore, a new HOG-LBP feature vector is proposed to express the logo characteristics including the rich texture details and the complex edge. With this change, recognition rate of 826 logo images belonging to ten distinctive vehicle manufacturers is improved by 2% than single HOG or LBP feature after the feature vectors fed to SVM classifier. We also analyze and optimize the vehicle faces’ HOG parameters to promote the recognition. The experimental results about 225 images including eighteen vehicle categories demonstate that non-overlapping blocks sliding the window of HOG feature has the better recognition results.Finally, based on the DM8168 DSP hardware and the DVR RDK software platform, collection and network transmission of HD video images are realized. Thus, a program is developed by VS2010 software and Open CV package and running in a backstage server to actualize the above-mentioned algorithms. In addition, motion detection algorithm is transplanted into the DM8168 DSP to transmit single vehicle image only from the front-end DM8168 to the server. This program, whose name is DSPALG Link, makes the network load lower. The experimental results indicate that this link can achieve the real-time extraction of video sequences.