prague12

In this thesis, an innovation for detecting the nonoccupied car parking space proactively on the roadside based on the vehicle video recorder was proposed. The vehicle video recorder was embedded an image recognition function to detect the parking space on the road side while the vehicle is moving along the road. When a nonoccupied parking space is detected, the image and its GPS information are sent to the backend system via wireless communication system for updating the available parking space data on the database. To improve the recognition rate of the vehicle video recorder, the classifier on the backend is retrained periodically based on the received samples from the end users, and then using the retraining result to update the classifier on the vehicle. To encourage the user feedback the positive parking space information to the system for accumulating the training samples, the system is operated with membership and nonmembership. In this research, a bi-block size double hits (BBDH) was proposed. To increase the probability for finding the objects, and resulting to reduce the processing time per frame. The classifier used is a combination of Adaboost and the probabilistic boosting tree algorithms to improve the system’s recognition rate. According to the experimental results shown,the proposed method can improve the detection rate by 3~12% compared to the existed methods. The maximum recognition rate could reach 88% underthe processing rate is 20 frames per second. The figures had shown that the proposed system is suitable for real applications.