A Small Car Model Recognition Embedded System

Car planet recognition techniques have been developed in the last decades. Most proposed systems are based on PC platforms to reach the real-time processing purpose and usually used in a video-based surveillance system or in a parking lot management system. To enhance those systems, a small-car-model recognition embedded system using Ada-boosting algorithm is proposed. Ada-boosting algorithm has been widely used in real-time face detection and recognition applications in the last 5 years. Its idea is simple by using some simple features to generate weak classifiers and a strong classifier is formed by these weak classifiers. Furthermore, each of car manufactures has its own design such as car shapes, rear light shapes and the component locations. These differences intimate that car models can be recognized by difference information, especial to small cars. To an embedded system, many traditional image processing techniques are not useful as in a PC platform because of some hardware limitations according our experiences of using an S3c2410 microprocessor in ARM920T architecture. It is a challenge to find out the necessary processing techniques and algorithms to explore the limited hardware abilities. In this embedded system to reach the real-time purpose, the minimum useful features of weak classifiers for the small-car-model recognition will be investigated. This proposed system will enhance the security of a car-planet recognition system with a car central database such as the one in Department of Motor Vehicle to detect a fake car planet or to search the possible target in a surveillance system.

Project ID:PB9503-0061
External Project ID:NSC95-3113-P182-001