基于立体视觉的三维空间入侵检测

针对平面视频监控无法感知运动物体深度的缺点,提出了基于立体视觉进行三维空间入侵检测的理论和方法。采用双目平行摄像系统,基于立体校正后的左右图像进行视差计算,并在视差图的基础上直接对运动物体做入侵判别;同时,设计了三维敏感区域的设置和报警策略。为了加快计算速度,采用一种基于下采样块匹配的方法来计算代价空间,最后用快速的局部动态规划输出视差图。在计算某目标的实际视差时,与目标的运动区域检测相结合,以减小计算误差。实际测试结果表明,该方法可以有效地探测到三维空间的运动目标入侵。

Intrusion Detection in Three Dimensional Space Based on Stereo Vision

Targeting at the inability of the depth of moving objects to be perceived in the commonly used "plane video" surveillance system,the theory and method of intrusion detection based on 3D vision and a fast calculation method of disparity are proposed.The disparity is computed based on the rectified stereo image pairs,which are captured from a parallel binocular camera system.The intrusion of moving objects is identified directly based on the disparity map,according to the 3D sensitive region settings and alarm strategy.In order to speed up the disparity computation,the cost space is calculated using a method of down-sampled block matching,and a fast local dynamic programming algorithm.In the disparity calculation of an object,it is combined with the moving region of the object in order to reduce the calculation error.The actual test results show that the method can effectively detect the intrusion of moving objects in 3D space.