高级检索
公绪超, 李宗民, 李冠霖, 李华. 基于区域判别与信息反传的车辆精细定位方法[J]. 计算机辅助设计与图形学学报, 2019, 31(12): 2136-2143. DOI: 10.3724/SP.J.1089.2019.17456
引用本文: 公绪超, 李宗民, 李冠霖, 李华. 基于区域判别与信息反传的车辆精细定位方法[J]. 计算机辅助设计与图形学学报, 2019, 31(12): 2136-2143. DOI: 10.3724/SP.J.1089.2019.17456
Gong Xuchao, Li Zongmin, Li Guanlin, Li Hua. A Precise Vehicle Location Method Based on Regional Discrimination and Reverse Information Transmission[J]. Journal of Computer-Aided Design & Computer Graphics, 2019, 31(12): 2136-2143. DOI: 10.3724/SP.J.1089.2019.17456
Citation: Gong Xuchao, Li Zongmin, Li Guanlin, Li Hua. A Precise Vehicle Location Method Based on Regional Discrimination and Reverse Information Transmission[J]. Journal of Computer-Aided Design & Computer Graphics, 2019, 31(12): 2136-2143. DOI: 10.3724/SP.J.1089.2019.17456

基于区域判别与信息反传的车辆精细定位方法

A Precise Vehicle Location Method Based on Regional Discrimination and Reverse Information Transmission

  • 摘要: 伴随智能交通图片数据爆炸式增长,交通大数据车辆精确定位管理需求增加.针对SSD(single shot multibox detector)检测算法对小目标定位不稳定与距离弥散导致的车辆定位偏差问题,提出了一种端到端的车辆及内部细节定位方法.包括引入无效目标过滤机制,以有效地抑制定位偏差;添加区域位置信息反传层,共享低卷积层局部特征,扩大小目标局部感受野,有效地进行车辆内部细节定位.采用PASCAL VOC2007数据集进行的实验结果表明,文中方法达到了良好的效果.

     

    Abstract: With the explosive growth of intelligent traffic image data,the demand for accurate location management of large data vehicles has increased.In this paper,an end to end vehicle and internal detail position method is proposed to solve the problem of vehicle location deviation which caused by small target location instability and distance dispersion by SSD detection algorithm.Firstly we introduce invalid target filtering mechanism to effectively restrain location deviation.Secondly we add area location information back layer and sharing the local feature of low convolution layers.Thirdly we enlarge the size of target local receptive field and effectively locate the details inside the vehicles.From the PASCAL VOC2007 experimental results,the improved method has achieved good results.

     

/

返回文章
返回