基于主运动分析的野外视觉侦察系统——运动目标检测、跟踪及全景图的生成

A VISUAL SURVEILLANCE SYSTEM BASED ON PRINCIPAL MOTION ANALYSIS FOR OPEN COUNTRY ——MOTION TARGET DETECTION, TRACKING AND PANORAMA GENERATION

本文介绍了一个针对野外自然环境设计的视觉侦察系统 ,其主要功能包括运动目标的检测、跟踪及环境全景图的生成 .摄像机安装在 36 0°水平旋转可控的车载云台上 ,视频信号及云台方位数据通过无线通讯传送给基地端的计算机 .计算机自动地检测其中的运动目标 ,实时地拼接出镶嵌有运动目标的全景图 ,当用户指定某个运动目标后计算机将自动地控制云台跟踪该目标 .我们根据二维仿射运动模型和鲁棒参数估计方法分析帧间主运动 ,对各帧图像进行运动补偿后求得差值图像 ,聚类分割变化点得到运动目标 ,再根据目标运动的连续性约束进行可靠的跟踪 .为了满足实时处理的要求 ,程序进行了优化 ,该系统在 PC机上 ,在 384× 2 88图像分辨率下达到 12 .5帧 /秒的处理速度 .

In this paper, a visual surveillance system for open country is introduced, which is composed of motion target detection, tracking and panorama generation. A camera is mounted on a 360° controllable platform for a vehicle with its video signal and orientation data transferred to the computer at remote base via wireless communication channels. The computer automatically detects motion targets, and generates panorama with motion target mosaics in realtime. When a target is selected by an user, the computer will control the camera platform to track it. The affine motion model and robust parameter estimation method are used to analyse principal inter-frame motion which will be used for motion compensation before calculating difference picture to locate motion target via clustering. Then target is tracked robustly based on motion smoothness constraint. Programs are optimized which can run on a PC for 384×288 size of processed image with a speed of 12.5 Hz.