摘要
针对巡检机器人检测异物范围受限,只能检测经过训练的异常物体的问题,提出了一种基于视频对齐的背景差分技术。巡检机器人由于运行速度不稳定、摄像头角度存在偏差,会导致视频无法对齐,识别准确率低,因此采用一种改进的联合优化算法对视频进行处理。首先使用改进的DTW算法逐帧查找与目标视频对应的帧,将视频对应帧提取出来;再采用SURF算法提取对应帧的特征点,进行筛选后使用DeepFlow算法得到对应帧图像的变形场矩阵,进而将图像角度修正,并进行像素级的匹配。实验结果表明,DTWSURFDeepFlow联合算法可以将图像进行完全的对齐,经过对齐的图像通过背景差分即可以检测出任意的异常物体,提高了异物检测的准确性。
Aiming at the problem that the inspection robot has limited detection range of foreign objects and can only detect abnormal objects after training,a new background difference technology based on video alignment is proposed.Due to the unstable running speed of the inspection robot and the deviation of the camera angle,the video cannot be aligned and the recognition accuracy is low.Therefore,an improved joint optimization algorithm is used to process the video.First,the improved DTW algorithm is used to find the frame corresponding to the target video frame by frame,and the corresponding frame of the video is extracted;then use the SURF algorithm is used to extract the feature points of the corresponding frame,and after filtering,the DeepFlow algorithm to obtain the deformed field matrix of the corresponding frame image,and then the image angle is corrected and pixellevel matching is performed.Experimental results show that the DTWSURFDeepFlow joint algorithm can completely align the images,and the aligned images can detect any abnormal objects through the background difference,which improves the accuracy of foreign object detection.
作者
杨涛
李祎
陈晶华
文炜
张香怡
顾继俊
YANG Tao;LI Yi;CHEN Jinghua;WEN Wei;ZHANG Xiangyi;GU Jijun(CNOOC(China)Co.,Ltd.,Beijing 100010,China;Sinopec Sichuan East Gas Pipeline Co.,Ltd.,Wuhan 430020,China;College of Mechanical and Storage Engineering,China University of Petroleum(Beijing),Beijing 102200,China)
出处
《机械与电子》
2020年第12期60-64,共5页
Machinery & Electronics
基金
天然气管道智能视频监控系统研究系统数据研发(3515057320ZC06070002)
国家重点研发计划项目(2016YFC0303706,2017YFC0805803)。
关键词
巡检机器人
视频对齐
像素级匹配
背景差分
异物检测
inspection robot
video alignment
pixellevel matching
background difference
foreign object detection
