摘要
针对运动车辆阴影带来车辆合并及形状失真的问题,提出一种基于Fast RCNN模型的车辆阴影检测去除算法。采用Selective search法对视频车辆图像提取多个车辆候选目标矩形区域,采用Hessenberg分解法将运动车辆和其阴影区域分开;利用深度网络提取阴影特征,用PCA分析法检测阴影,训练优化该网络,识别移动阴影中包含的车辆区域,实现快速去除阴影的效果。实验结果表明,该方法有效解决传统算法多车辆阴影检测去除效率低下问题,平均检测精度mAP(mean average precision)提高2.78%,为智能交通系统提供良好技术基础。
Aiming at the problem of vehicle merging and shape distortion caused by moving vehicle shadow,a vehicle shadow removal algorithm based on Fast RCNN model was proposed.The selective search method was used to extract the rectangular area of multiple candidate targets in the video vehicle image.The moving vehicle and its shadow area were separated using Hessenberg decomposition method.The shadow feature was extracted from the depth network,and the shadow was detected by PCA analysis.The network was trained and optimized.The vehicle area contained in moving shadow was identified to achieve the effect of fast shadow removal.Experimental results show that the proposed method can effectively solve the low efficiency problem of the traditional detection algorithm for vehicle shadow removal,the average detection accuracy of mAP（mean average precision）increases by 2.78%,providing agood technical foundation for the intelligent transportation system.
出处
《计算机工程与设计》
北大核心
2018年第3期819-823,共5页
Computer Engineering and Design
