胡涛, 朱欣焰, 呙维, 张发明. 融合颜色和深度信息的运动目标提取方法[J]. 武汉大学学报 ( 信息科学版), 2019, 44(2): 276-282. DOI: 10.13203/j.whugis20160535
引用本文: 胡涛, 朱欣焰, 呙维, 张发明. 融合颜色和深度信息的运动目标提取方法[J]. 武汉大学学报 ( 信息科学版), 2019, 44(2): 276-282. DOI: 10.13203/j.whugis20160535
HU Tao, ZHU Xinyan, GUO Wei, ZHANG Faming. A Moving Object Detection Method Combining Color and Depth data[J]. Geomatics and Information Science of Wuhan University, 2019, 44(2): 276-282. DOI: 10.13203/j.whugis20160535
Citation: HU Tao, ZHU Xinyan, GUO Wei, ZHANG Faming. A Moving Object Detection Method Combining Color and Depth data[J]. Geomatics and Information Science of Wuhan University, 2019, 44(2): 276-282. DOI: 10.13203/j.whugis20160535

融合颜色和深度信息的运动目标提取方法

A Moving Object Detection Method Combining Color and Depth data

  • 摘要: 行人检测是计算机视觉、智能交通等领域研究的热点与难点,基于深度传感器对室内复杂场景下的行人检测展开研究。目前,基于颜色与深度数据的目标检测方法主要包括基于背景学习的方法和基于特征检测算子的方法,前者依赖于视频序列头几十帧的背景知识,帧的数量决定检测质量;后者存在计算量大的问题,训练样本的不足也会影响行人检测结果。因此,深入分析了复杂场景特征,融合颜色和深度信息,提出了RGBD+ViBe(visual background extractor)背景剔除方法,实现前景运动目标的准确提取。实验结果表明,提出的RGBD+ViBe方法在前景运动目标检测准确率方面要明显高于仅考虑颜色或深度信息方法以及RGBD+MoG(model of Gaussian)方法。

     

    Abstract: Pedestrian detection is a hot topic and difficult problem in areas like computer vision and intelligent traffic. This paper researches on pedestrian detection in complex indoor space using depth sensors. In recent years, target detection methods based on RGB-Depth(RGBD) data mainly include background learning method and feature detecting operator method. However, the former method depends on the background knowledge of first tens of frames, and the number of frames decides the final detection accuracy. The latter takes plenty of time for computing, and being lacks of training samples may influence the detection result. Thus, this paper analyzes the complex scene features and integrates the color and depth information, and proposes a RGBD+ViBe (visual background extractor) background elimination method. The experiment results indicate that the detection accuracy of the proposed method is higher than the methods which only consider color or depth information and the RGBD+MoG method in foreground extraction.

     

/

返回文章
返回