道路监控视频中运动目标分割算法的研究

Study of Moving Object Segmentation Algorithm in Traffic Surveillance Video

运动目标分割是交通事件检测的基础,分割的质量直接影响道路事件检测的准确性.依次研究了平均法、连续帧差法、混合高斯建模及阴影消除等算法,并综合应用于道路监控视频的运动目标分割的各阶段.针对连续帧差法和混合高斯建模算法进行了仿真实验,结果表明,混合高斯建模算法更适合做运动目标的分割,但由此得到的前景运动目标常受到阴影干扰,所以在混合高斯建模算法中,加入了消除运动目标阴影的算法.通过实验仿真表明,优化的该算法能够更加清晰地分割出前景运动目标图像.

The moving object segmentation is the basis for traffic incident detection.The quality of segmentation algorithm directly influences the accuracy of traffic incident detection.The average method,the consecutive frame difference method,the Gaussian mixture modeling algorithms and shadow elimination are studied here.These algorithms are integrated and applied in various stages of the moving object segmentation in traffic surveillance videos.In particular,the consecutive frame difference method and the Gaussian mixture modeling algorithm are taken into experiments.The results indicate that the Gaussian mixture modeling algorithm is more suitable for the moving object segmentation.However,the foreground moving object is often interfered by the shadow,so the algorithms of shadow removal are added to the Gaussian mixture modeling algorithm.The experimental results show that the optimized algorithm can more accurately segment the foreground moving object images.