基于DM642的运动目标检测

Moving target detection based on DM642

提出了一种高斯混合背景模型和YUV色度空间相结合的运动目标检测算法。高斯混合模型对背景光线变化有较强的鲁棒性,且对背景中的周期性变化有较好的抑制作用,检测出的目标有较好的连通性;但其对于全局亮度的变化及噪声较为敏感,容易误判。为此选取对亮度变化不敏感的UV分量来进行运动目标检测,然后再和Y分量的高斯混合背景检测进行"与"运算,从而消除高斯模型的误检,最后针对运动目标的影子问题,采用基于垂直投影图的阴影消除算法除去影子。算法在DM642开发板上实现。实验结果表明,该算法能够实时精确地检测出运动目标,且对全局光照变化不敏感。

Proposes an efficient moving target detection algorithm, which combines the GMM with the YUV color space. The GMM has strong robustness to the change of the light in the background, and it also has a good inhibitory action to the periodic changes of the background. The detected target has good connectivity. However, the GMM is sensitive to global noise, global brightness change, and easy to mistakenly judge the background points as the foreground points. In order to eliminate the GMM＇s false alarms, uses the UV components, which are not sensitive to the changes of the global brightness, for the motion detection, and then the result will do ＇AND operation＇ with the Y compo- nents motion detection result. Finally, uses the algorithm based on vertical projection shadow elimina- tion to remove the shadow of the motion target. The algorithm is implemented on the DM642 develop- ment board. Experimental results show that the algorithm can accurately detects the moving target in re- al-time, and it is not sensitive to the global illumination changes.