基于多特征融合的运动车辆阴影去除算法-《郑州大学学报(工学版)》

文章信息/Info

Title:: Shadow removal algorithm for moving vehicles based on Multi-feature Fusion

Author(s):: Wu Guodong; Zhang Aimei; Huang Xiao; Yao Pengwei; School of Mechanical Engineering, Zhengzhou University

摘要:: 在交通监控视频中，需要检测运动车辆的信息，而车辆的阴影常常被误检测为车辆本身，直接影响车辆检测的准确度，为后面的技术处理带来困难。本文应用一种基于颜色空间、纹理特征及边缘特征的多特征相结合车辆阴影去除方法。首先，通过传统的混合高斯方法建立背景模型，并提取前景目标。其次将前景目标在HSI（Hue-Saturation-Intensity）空间通过阈值法检测阴影，由于基于颜色模型的检测易将运动目标本身误检测为阴影，采用LBP算子结合边缘特征检测方法提取运动目标。最后将LBP算子和边缘特征检测出的前景目标与HSI颜色空间检测出的阴影目标相结合就可以检测出实际阴影区域。将检测出的阴影区域去除即可得到真实的运动前景目标。仿真实验的结果表明，本文算法可以有效去除交通监控视频中运动车辆的阴影，在运动车辆阴影检测率上提升约10%。

Abstract:: In traffic surveillance video, it is necessary to detect the information of moving vehicles, however, the shadow of the vehicle is often wrongly detected as the vehicle itself, which directly affects the accuracy of the vehicle detection and brings difficulties to the technical processing behind the vehicle detection .In this paper, a vehicle shadow removal method based on color space, texture feature and edge feature is proposed. First, the background model is established by the traditional mixed gaussian method, and the foreground target is extracted.Second The shadow of the foreground target is detected by threshold method in HSI (Hue-Saturation-Intensity) space,because the detection based on color model is easy to detect the moving object itself as shadow, the LBP operator combined with edge feature detection method is used to extract the moving target.Finally, combining the foreground target detected by LBP operator and edge feature with the shadow object detected in HSI color space, the actual shadow region can be detected.The real moving foreground target can be obtained by removing the detected shadow area. The simulation results show that the proposed algorithm can effectively remove the shadow of moving vehicle in traffic surveillance video, and improve the detection rate of moving vehicle shadow by about 10%