面向监控场景的运动目标分割算法

CDnet2012研究数据为CVPR2012年举办的Workshop的一项竞赛数据，该数据集由31个真实世界的视频（包括红外视频）组成，总计超过80，000帧，包括监控场景下各种运动和变化检测挑战。该数据集代表了当今在监控和智能环境场景中捕获的室内和室外视觉数据，并包括一套全面的人工标注的实况变化以及运动区域的标签。 CDnet2012运动目标检测与分割实验结果数据包含基于高斯尺度混合模型刻画稀疏前景运动目标估计的不确定性以及动态复杂的低秩背景中稀疏的运动干扰项方案下的相关实验分割结果数据，即对研究数据集CDnet2012处理后的运动目标分割图，共包含31个真实世界的视频的处理结果。处理结果总计超过65,000帧，包括对于监控场景下各种运动与变化检测分割挑战的处理结果。

The CDnet2012 research dataset is a competition dataset from the workshop held in conjunction with CVPR 2012. It comprises 31 real-world videos, including infrared videos, with a total of over 80,000 frames, and covers various motion and change detection challenges in surveillance scenarios. This dataset encompasses indoor and outdoor visual data captured in contemporary surveillance and smart environment settings, and includes a comprehensive set of manually annotated ground-truth change labels and motion region labels. The CDnet2012 motion object detection and segmentation experimental result dataset contains relevant experimental segmentation results derived from the approach that uses the Gaussian Scale Mixture model to characterize the uncertainty of sparse foreground motion object estimation and sparse motion disturbances in dynamic complex low-rank backgrounds. Specifically, these are the motion object segmentation maps generated by processing the CDnet2012 research dataset, covering the processing results of 31 real-world videos with a total of over 65,000 frames, including the results addressing various motion and change detection and segmentation challenges in surveillance scenarios.

搜集汇总

背景与挑战

背景概述

该数据集基于CDnet2012竞赛数据，由31个真实世界监控视频组成，总计超过80,000帧，包含室内外场景的人工标注运动区域。同时，它提供了基于高斯尺度混合模型处理后的运动目标分割结果，涵盖超过65,000帧，用于支持监控场景下的运动与变化检测研究。