摘要
针对汽车的追尾碰撞事故,提出了基于OpenCV的前方车辆检测和多信息融合预警的方法。该方法首先利用Haar-like+Gentle Adaboost实现前方车辆的快速识别,结合Kalman滤波原理跟踪车辆,实现前方车辆检测,然后基于几何模型实时计算前车与本车的横纵向距离,最后根据本车及前车车速、碰撞时间TTC、横向距离等信息与阈值进行比较,分级识别碰撞风险。试验结果表明,该检测方法平均耗时22 ms/帧,检测率达到96%,并能较准确地测量车距,实现可靠的前方避撞预警输出。
A preceding vehicle detection and multi-information fusion warning method based on OpenCV was proposed for the rear-end collision accident, that utilized Haar-like＋Gentle Adaboost to detect preceding vehicles rapidly, and used the Kalman filter principle to track these vehicles. Geometric model was used to calculate the lateral and longitudinal distances in real-time with the preceding vehicle, then vehicle speed and the preceding vehicle speed, collision time 33℃ and lateral distance, etc., were compared with threshold value to identify the risk of collision. The experimental results show that the proposed detection method can detect preceding vehicles in about 22. ms per frame with an accuracy of 96%, and can measure the vehicle distance accurately to realize reliable front collision warning output.
出处
《汽车技术》
CSCD
北大核心
2017年第6期11-16,共6页
Automobile Technology
基金
江苏省高校自然科学研究重大项目(13KJA580001)
关键词
前方车辆检测
碰撞预警
KALMAN滤波
多信息融合
Preceding vehicle detection, Collision warning, Kalman filter, Multi-informationfusion
