复杂场景双目立体视觉测量精确定位系统研究  

Research of binocular stereo vision measurement system for complex background

作　　者：龙邹荣 冯鹏[1] 杨荣来 任勇[1] 刘远远 魏彪[1] 

机构地区：[1]重庆大学光电技术及系统教育部重点实验室,重庆400044

出　　处：《光电子．激光》2017年第11期1252-1259,共8页Journal of Optoelectronics·Laser

基　　金：DIRLS-001图像识别定位系统(JG2015068)资助项目

摘　　要：针对某复杂场景下对特定目标进行精确定位测量的关键问题,依据双目立体视觉测量原理,结合OpenCV视觉开发库标定摄像机内外参数,利用结果校准目标图像,借助于标靶类型进行图像区域分割并计算出目标位置坐标,以此实现精确定位测量之目的。研究结果表明,对于所确定的1.4~1.6m目标测距范围,采用平行双目视觉测量模型和相机小孔测量模型,获得了距离误差小于10mm、横向误差小于2 mm和响应时间小于350ms的测量结果,在满足实时性的同时具有较高的定位精度。Aiming at the key problems of precise positioning and measurement of objects or objects in scenes, this paper designs a measurement system based on binocular stereo vision and open source com- puter vision library （OpenCV）. This project aims to achieve real-time positioning of the target. The sys- tem calibrates 5 million industrial CCD cameras with the method provided by OpenCV. Then the calibra- tion data is used to calibrate the target images which are used to achieve regional segmentation images. The distortion caused by the lens process can be removed by calibration. Image parameters can also be consistent. We use the segmentation algorithm to identify the target in the image. Finally, we get the tar- get coordinates from segmentation data, with which the camera＇s spatial attitude can be calculated through the camera geometry. The results show the system has superior comprehensive properties. The experimental measurements show that the results are excellent in the range of 1.4-1.6 m. Response time is less than 350 ms,distance error is less than 10 mm and transverse error is less than 2 ram. The test results meet the requirement of the project indicators.

关 键 词：双目立体视觉测量 精确定位 CCD摄像机 标靶 OPENCV 

分 类 号：TH741[机械工程—光学工程]