视频图像中监控目标的空间定位方法

Estimation of target position from video image

为解决视频监控系统中目标空间定位存在困难大、耗时等问题，根据OpenGL透视成像与摄影测量一致性原理，提出基于三维场景仿真成像，并依靠视频图像与仿真图像问的对应关系进行地理场景中目标的空间定位方法．在视频成像时摄像机内外方位元素已知的情况下，在三维虚拟场景中形成虚拟相机来模拟实际摄像机成像过程，并将监控目标所在视频图像中的像素坐标等比代入仿真图像，通过虚拟成像的逆过程反推出目标的空间坐标．为有效分析系统的定位精度，采用地面激光扫描仪＋同步拍照相机组成的三维场景采集系统来模拟实际的高精度摄像机＋数字云台组成的视频监控系统．实验结果表明，目标点与相机的水平距离介于10—90m内，各目标点的空间定位在X、Y、Z三个方向的误差大多在±0．31m之间，说明本方法是可行的．这对于通过视频监控图像实现户外目标的定位，具有重要的参考价值．

Video surveillance technology, which provides intuitive, accurate, realtime and rich information, has been widely used in various fields. Generally, video surveillance system requires that human operator identifies the target and its However it is a difficult task to locate the natural geographical environment without position from the video image to assist decision -making. target position timely, even harder in large and complicated taking reference to the proper geographic positions. In order to determine and estimate the outdoor monitor target position from video images we propose a outdoor position estimation method which is based on virtual imaging simulation of three - dimensional scene and video image. This method is developed according to the consistency of the OpenGL persective ima- ging mechanism and photogrammetry theory. In the condition of knowning camera photogrammetry exterior and interior parameters, we firstly create virtual camera model for simulating actual imaging process in three - dimensional scene, then the screen coordinates are derived based on the target pixel coordinates in the single frame video image, and later tramsformed into world coordinate system ac- cording to the inverse process of virtual imaging. In order to effectively certify system of the positioning accuracy, we use terrestrial laser scanner and digital camera to obtain 3D point cloud data and digital images to simulating the actual video monitoring system composed of high precision digital holder and digital camera. The results show that the relative average error for estimation the monitor target spatial positioning relative is 1. 160%, and the error of X, Y, Z three directions is mostly between -0.3 m and 0.3 m. The distance between target and camera range from 10 to 90 m. So the target spatial positioning method is feasible. The method will be useful for estimating outdoor moving target positon in wide range of geographical scene by video surveillance system.