中文摘要：

提出了一种基于车辆行为识别的汽车前方碰撞预警方法.利用单目视觉,首先采用基于梯度方向直方图特征和支持向量机的方法识别前方车辆,并结合卡尔曼滤波进行车辆跟踪;然后使用隐马尔科夫模型对车辆行为进行建模,识别前方车辆行为,并根据行为识别结果计算对应的风险评估因子;最后将风险评估因子引入碰撞风险评估系统,使碰撞预警时间比未加入风险评估因子平均提前2.04s.实车实验验证了本方法的有效性.

英文摘要：

A collision warning system was proposed based on the forward vehicle behavior recognition method.Based on the monocular vision,the support vector machine (SVM)classifier was applied based on histogram of oriented gradients (HOG)feature for vehicle detection,and the Kalman filter was applied for vehicle tracking.Then the hidden Markov model (HMM)was used for vehicle behav-ior recognition.According to the vehicle behavior recognition result,the corresponding risk assess-ment factor was calculated,which was considered in the risk assessment system,and the performance of the collision warning system was improved by 2.04 s compared without the corresponding risk as-sessment factor.Finally,the effectiveness of this forward vehicle behaviors recognition system was verified using real-word on-road data.