Research On Vehicle Logo Recognition

The vehicle logo recognition technology is a digital image processing and pattern recognition application that automatically identify the logo of vehicle from a digital image or a video frame from a video source, which is a important aspect of the intelligent transportation system.It is applied in many fields such as highway toll, traffic administration and control, electronic police, etc. The vehicle logo is very important discriminative information for vehicle brand classification. Vehicle logo location and recognition are two critical issues of the logo recognition system. The special characters such as small dimension and complex background make the logo image sensitive to light and imaging angle. The key point of the recognition system is to find the suitable feature descriptors that must be robust to geometric distortion, uneven illumination, partial defect and noise.A new vehicle logo recognition algorithm is described in this thesis. The algorithm consists of the location module and the recognition module. What makes this algorithm remarkably different from the other methods is the application of prior knowledge. The key technologies of logo recognition are studied and an effective method is proposed. The vehicle logo recognition simulation software based on the proposed algorithm is developed and experiments are conducted.In the vehicle logo location stage, a prior knowledge-based coarse-to-fine method is proposed. The main steps are: 1) using the background updating method to detect candidate vehicles; 2) using the orientation filter to locate car body; 3) using gradient projection to locate vehicle lamp zone; 4) using the layout of the vehicle lamp zone to roughly locate the vehicle logo region; 5) using edge extraction method to locate the fine logo region.In the vehicle logo recognition stage the width-height ratio is used to construct the first-order classifier, then the SIFT descriptors are extracted as the features which enter the BP neural network for logo recognition.The experimental results show that in the vehicle logo recognition system, the location rate is about 95.95%, the recognition rate is about 74.77% and the error recognition rate is about 15.51% and rejection rate is about 9.72%.