摘要
车辆目标的粘连检测和分割是实现交通场景中目标跟踪和路面车流量统计等功能必须解决的关键问题.首先,采用高斯降采样滤波处理剔除车体细节噪声干扰,利用镜像对称LBP编码的相似性对车体对称结构进行分析,提取对称轴;然后,利用对称轴分布的数量和位置信息进行车辆目标粘连检测;最后,结合车体对称结构信息和目标轮廓凹性分析给出车辆轮廓最优分割方案.实验结果表明,在满足实时处理速度要求的前提下,相比传统的基于轮廓形状的分析算法,本算法在复杂的环境噪声干扰下具有更好的噪声适应性,并且在车辆粘连检测准确率和粘连分割能力方面均有更好的表现.
Adhesion detection and segmentation of vehicle targets is a key problem that must be solved to realize the function of target tracking and traffic flow statistics in traffic scenes. In this paper, Gaussian decampling filter was used to eliminate vehicle detail noise, the symmetry axis of the vehicle body was extracted by analyzing the similarity of the mirror symmetry LBP coding. And then, the number and position information of symmetry axis distribution were used to detect the adhesion of vehicle targets. Finally, the optimal segmentation scheme of vehicle contour was given based on the symmetric structure information of vehicle body and the concavity analysis of target contour. The experimental results show that, on the premise of meeting the real-time processing speed requirements, the proposed algorithm has better noise adaptability under the interference of complex environmental noises compared with the traditional analysis algorithm based on contour shape, and has better performance in terms of vehicle adhesion detection accuracy and adhesion segmentation ability.
作者
马秀博
孙熊伟
MA Xiubo;SUN Xiongwei(Department of Computer Engineering,Anhui Sanlian University,Hefei 230601,China;Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230088,China)
出处
《湖北大学学报（自然科学版）》
CAS
2022年第6期742-749,共8页
Journal of Hubei University：Natural Science
基金
安徽省教育厅自然科学重点项目(KJ2017A524)
校级自然科学基金重点项目(KJZD2020008)资助。
关键词
对称检测
LBP编码
遮挡检测
粘连分割
symmetry detection
LBP
occluded vehicles detection
adhesion segmentation
