The Anomaly Activity Detection Method Of Mine Video Based On Graph Structure Matching

At present,the research on videos anomaly detection has become a hot issue of increasing concern to society and government.In the field of mine,because of the harsh environment,many kinds of noises and poor quality of'video data,it has seriously affected the recognition of abnormal activities and the description of normal video semantics,which is not conducive to intelligent monitoring of safe production in coal mine.The hidden danger of coal mine safety is still serious.Therefore,it is necessary to conduct in depth research on video abnormal activity detection of complex scenes in coal mine to provide theoretical support and technical support f'or intelligent monitoring and warning in coal mine.In the detection of abnormal activity of underground video in coal mine.First of all,we need to use key frame extraction algorithm to characterize the semantic elements of underground video data by using a small number of key frames.Therefore,this paper proposes a method of extracting key frames based on secondary filtering of adaptive Top-K,aiming at using an adaptive threshold as the K value of Top-K to extract key frames.Firstly,the video frame is extracted from the mine video,and the moving object is extracted from the differential background.Secondly,SIFT is used to extract the features of moving objects and fit the video feature curve.Then,adaptive Top-K algorithm is used to find out the eigenvalues which are larger than an adaptive threshold.Finally,according to the adaptive threshold,the feature values are filtered twice by Top-K alg,orithm.and the key-frames of video images are obtained.The experiimental results show that the proposed method can extract the key frames of underground video effectively and accurately.Wich provides better data support for further detection of abnormal video activity.Based on secondary filtering of adaptive Top-K key frames extraction method,this paper proposes a method of abnormal activity detection for mine video based on graph structure matching,which aims to match the effective features depicted by depth neural network with graph structure method for further classification and recognition.Firstly,the VGG16+LSTM deep learning framework is used to extract the features of video key frames.The extracted features are combined with the spatial features of VGG16 neural network and the temporal features of LSTM neural network.Secondly,the graph structure method is used to match the maximum connected graph in the feature of the neural network as an effective feature to replace the original feature of the neural network.Finally,the video frames with abnormal activity are obtained by SVM classifier.The validity and robustness of this method are verified on the data set of underground coal mine.And compared with the traditional SIFT feature extraction method and the existing CNN+LSTM neural network method,the results show that this method can still detect abnormal activities in underground coal mine scene with complex background and changing illumination.Compared with other abnormal activities detection methods,the performance of this method is improved.