Research On Joint Detection And Recognition Of Vehicle Logo

The vehicle logo detection and recognision play an important role in Inteligent Trans Portation System, vehicle tracking and vehicle information management. Since various types and shapes of the vehicle logo, uneven illumination, edge blur and complex background of vhicle logo image in the application, it is difficult to accurately detect vehicle logo area, adapt to various types and multi-scale. To solve these problems, research in this paper includes texture feature extraction,standard multi-class and multi-scale database building, vehicle logo anti-interpretation,etc. The contents are shown as follows.1. To solve the difficulty of extracting enough information in small-size logo’s texture feature, using the Dense SIFT algorithm by extracting pixel and its neighborhood gradient information.2. To build multi-class and multi-scale standard vehicle logo off-line database, we had selected typical standard vehicle logo templates, extracted the multi-scale features.3. To locate the vehicle logo area accurately,Proposing an improved SIFT Flow vehicle logo anti-interpretation algorithm. Forming the energy function, including four hypothesis, multi-scale term, anti-interpretation term, similar SIFT features of the same logos.4. To detect and recognize multi-class and multi-scale vehicle logos, we proposed a vehicle logo joint detection and recognition algorithm. This paper proposed a decision function based on maximum a posteriori estimation, including 3 cost function, mean energy term, interpretation similarity term, etc., converging the detection and location module.5. Vehicle logo detection and recognition software based on the video stream, application of parallel acceleration and hybrid programming to improve algorithm’s speed.