Logo Recognition Technology Research Vehicle Based On The Image Of Intelligent Transportation Systems

Inteligent Transportation Sysytem has become a trend of traffic development this century, and vehicle-logo recognition system is an important part of ITS, so researching upon vehicle-log recognition has the practical significance. The image-based vehicle-logo recognition system includes several parts, such as image capture, vehicle-log detection, vehicle-log recognition and application, among which vehicle-log detection and recognition are the two key technologies. So, this paper mainly studies the detection and recognition technologies. The main content as follows:(1) This paper studies a creative vehicle-log detection method by visual saliency. The target detection based on visual saliency has been a novel method resent years. The typical salient mode Itti is improved by extracting the colour features、intensity features、orientation features、edge features, which are linearly fused into the final salient map. Then operate binary segmentation、morphology closed operation、projecting, etc to detect the vehicle-log. This method is adaptable to different situations and it is robust.(2) This paper studies a vehicle-log detection measure by texture suppression. Firstly, this paper proposes a straightforward measure to classify the texture around vehicle-log into two categories:horizon-like texture and vertical texture, and then use different detecting algorithm respectively,which will suppress the texture firstly and then detect vechile-log by projecting. Experimental results demonstrate that this method has a better effect.(3) This paper extracts LBP patterns feature for vehicle-log recognizing. This paper studies the dimension reducing of LBP uniform patterns and LBP sub-patterns by SVM upon feature dimension、recognition time、accuracy, etc, which has a certain reference value. LBP uniform patterns is more efficient, and the LBP sub-patterns combined with PCA is more flexible in dimension reducing.(4) This paper proposes an improved KNN classification algorithm (Namely MKNN). The proposed algorithm combines the advantage of MDC and KNN. Firstly, sort the distance between the sample and each classification center according to MDC in ascending order and get the top M classifications, then get the concrete classification according to KNN among the M classifications. It is tested upon large data sets. The experimental results demonstrate that MKNN is efficient and it also has a high accuracy.(5) This paper uses KNN、MKNN、SVM methods to classify vehicle-log, and analyses the different identification results and reasons of errors. In addition, this paper designs a simple MFC interface to display the results of the core algorithms.