视频分析技术在日志系统中的应用

近年来,以多传感器信息融合为特征的日志系统成为了一个新的热点问题。日志系统使用者携带便携式摄像头所获得的视频信息是一类主要信息,获取其关键帧序列对完善系统功能具有重要意义。与其他应用相比,日志系统视频更加复杂多变。基于以上条件,本文提出非参数模型的方法估计背景运动;针对前景运动物体检测和跟踪问题,本文在非监督模式识别方法的基础上,使用统计、时空、颜色特征等给出前景区域判断;最后,利用上述结果得到关键帧序列。

Application of Video Analysis Technologies in Life-log System

In recent years, Life-Log system, which is characterized by fusion of information from Multi-Sensors, has become a new research hotspot. Video captured by wearable-camera plays the most important role in the Life-Log system, so if applying video analysis technologies in Life-Log system, it would be possible to improve the functions of the system. Comparing with other applica-tions, the changing of the scene in life-log system is more complex. Considering the constraints, this paper purposes a new method called non-features-model method to estimate the motion of video. Considering the problem of foreground moving object detection, this paper uses unsupervised learning algorithm to classify the motion vector field, and then use the color, spatial and statistics fea-tures to give the result of foreground object tracking. At last, the serial of key-frames is generated by the result of motion estimation and foreground object detection.