基于DBAN的监控视频数据挖掘

摘要

视频挖掘主要涉及三个层次的工作:视频数据预处理,视频特征数据提取及视频模式发现与表示.针对监控视频数据,以人体姿势识别和行为理解为挖掘任务开展视频数据挖掘研究.提出了一个带有二维身体部位表示法的动态贝叶斯动作网(DBAN)基本框架来提高人体姿势定位的准确性及行为识别的精度,并通过实验证明了该方法的有效性.%Video mining principally involves 3 levels of operation,namely,video data preprocessing,video feature data extraction,the discovery and representation of video mining pattern.Aiming at the above manipulations,the current study probe into the researches of video data mining by taking the recognition of body posture and the comprehension of action as the mining tasks.The paper puts forward a DBAN basic framework with the notation of planar body part so as to enhance the accuracy of body posture positioning and the precision of its action recognition,which has been given testimony to its availability through one empirical research.