遗传算法优化LVQ网络的监控视频关键帧内容识别  

Surveillance video key frame recognition based on LVQ network of genetic algorithm optimization

作　　者：马利克[1,2] 彭进业[1] 冯晓毅[1] 

机构地区：[1]西北工业大学电子信息学院,陕西西安710072 [2]陕西省人民警察培训学校,陕西西安710054

出　　处：《西北大学学报（自然科学版）》2015年第4期573-578,共6页Journal of Northwest University（Natural Science Edition）

基　　金：高等学校博士学科点专项科研基金资助项目(20116102110027);国家自然科学基金资助项目(61075014)

摘　　要：为解决监控视频检索中公安视频侦查关注目标的识别问题,提出一种基于遗传算法优化LVQ神经网络的关键帧内容识别方法。首先通过运动目标检测及二值图像的聚散熵,对监控视频进行子镜头划分,从而提取视频关键帧。其次归一化关键帧中的待识别目标,提取待识别目标的形状统计特征。再次构造LVQ网络并利用遗传算法对网络的初始权值进行优化,训练网络实现关键帧内容识别。最后列举出该方法的实验结果及性能分析。该方法在关键帧内容识别的准确性和鲁棒性上都有良好表现。To solve the problem of police concern target identification in surveillance video retrieval, a surveil- lance key frame recognition algorithm is proposed which based on an LVQ neural network of genetic algorithm optimization. Firstly, a moving object detection algorithm and the aggregation dispersion entropy were used to divide surveillance video into several shots. Thereby the key frame could be extracted. Secondlyl the object to be recognized was normalized, and a shape statistics feature was extracted. Thirdly, an LVQ neural network was established for key frame recognition. And the genetic algorithm was used to optimize the initial weights of neural network. Finally, the experimental results and discussions were given at the end of the article. This al- gorithm has good performance both in accuracy and robustness.

关 键 词：监控视频 关键帧 LVQ网络 遗传算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]