| With the development of intelligent video monitoring,more and more video monitoring equipment has been applied to various fields of security.Through video monitoring system,loopholes and deficiencies existing in medical services can be found,which can also provide effective evidence for public security organs to assist in solving cases.However,with the augmentation of monitoring equipment,monitoring the amount of video data is also increasing,traditional by the human eye to abnormal events in the video identification method requires a lot of manpower material resources,has been unable to meet the needs of today’s video analysis,so the intelligent target abnormal behavior based on the monitoring video detection system plays an important role.Based on the problems faced by the identification of abnormal target behaviors in video,this paper proposes a new detection model of abnormal target behaviors to intelligently identify and monitor abnormal human behaviors under video,including the following three aspects:1.A method of depth estimation based on monocular fixed camera is proposed.Aiming at the problem that it is difficult to obtain the human body depth information in 3d space directly in the existing methods,this paper proposes a target depth estimation method.By measuring the initial value,the depth information of the key nodes of the target skeleton is derived,and then the complete depth information of the human body in three-dimensional space is obtained.2.A framework motion recognition model based on spatiotemporal relation is proposed.Aiming at the problem that it is difficult to extract the key information in skeleton long time series by existing models,a skeleton motion recognition model based on space-time relation(LST-CNN& FS-LSTM)is proposed in this paper.Network improved the existing model of Local spatio temporal convolution neural network(LST-CNN)for three dimensional convolution in the image,according to each time period on the behavior criterion attention degree of different Fragment Selection Long Short-Term Memory(FS-LSTM),improve the LSTM network will output for an average of all the time to make the network more average to obtain all useful information,in order to achieve the goal of skeleton sequence period of choice.In addition to the problem of multi-person recognition of skeleton sequences this paper proposes a feature compression method based on variable pooling to dynamically process the behavior recognition of multi-person skeleton sequences.The validity of the model is verified by experimental comparison.3.The abnormal behavior detection system is implemented.In order to verify the effectiveness of the algorithm,this paper combined the demand development of the actual project,designed and completed the target abnormal behavior detection application system under the scene of the prison,and finally achieved the effect of abnormal monitoring in the prison. |
