| dc.description.abstract | In these decades, traffic accidents have increased year after year. For traffic safety, many companies and research institutes developed real-time vehicle safety systems. When driving on the road, the view fields beside the host vehicle are limited and the drivers can not clearly observe both sides of the blind spot area; then accidents may occur when the driver changes lane. In order to avoid the accidents, we use a camera mounted under a side-view mirror and detect the vehicles with computer-vision technology.
The proposed blind-spot detection system consists of seven stages: defining the detection zone, feature point detection and filtering, optical flow estimation, optical flow filtering, optical flow grouping, vehicle underneath shadow detection, and still object detection. The dynamic information can detect moving objects, but can not detect vehicles which are still relative to the host vehicle. Thus, we include static information to assist detecting these objects.
The proposed detection system has been test in various weather conditions including sunny, cloudy, rainy day, night, etc. In experiments, the detection rate is about 97% in sunny day, 95% in cloudy day and 92% in night. The performance of the proposed system reaches 30 frames per second on an Intel? Core 2 Duo? E8300 2.83 GHz CPU, 2GB DDR RAM running on Microsoft? Windows 7.
| en_US |