基于时空关系模型的城市道路车辆实时检测

复杂城市道路环境下运动车辆目标检测是现代智能交通系统的重要组成部分.依据多帧视频图像序列的时空连续变化关系，通过构建多帧视频图像序列时空关系模型（Time-space model——TSM），进一步完善车底阴影特征检测算法，并与AdaBoost算法相结合，实现运动车辆目标检测的候选区域筛选与验证处理，以降低车辆检测的误检率，提高准确率.在白天复杂城市道路环境下，实验结果显示基于TSM的车辆检测，检测准确率为92.1%，误检率为4.3%，图像分辨率为1920*1088，单帧图像平均处理时间76 ms.基于TSM的车辆检测显著改进了AdaBoost和车底阴影特征检测算法存在的误检率高，效率低问题，满足城市道路环境下车辆检测准确率和鲁棒性的要求.

Real-Time Urban Road Vehicle Detection Based on Time-Space Model

Urban road vehicle detection is an important part of modern intelligent transportation system-ITS. According to image sequence time-space relations of continuous change, the underneath shadow feature vehicle detection algorithm is further improved by constructing the time-space model of video image sequence, and is combined with the AdaBoost algorithm, filtering out false candidate region of vehicle. Experimental results demonstrate that the accuracy rate of proposed algorithm is 92.1%, with the false positive being 4.3%, the resolution of image being 1920*1088 and the time of processing being 76ms under the complex urban traffic environment. The algorithm effectively improves the high false detection rate and low effectiveness of AdaBoost and underneath shadow feature detection algorithms, and can meet the accuracy and robustness requirements of vehicle detection in the urban road environment.