| With the rapid development of the tourism market,abnormal behavior of some tourists often appears in tourist scenic spots.These behaviors not only endanger the safety of life,but also have a lot of negative effects on society.Therefore,it is of great significance to study the abnormal behavior recognition system of tourist scenic spots.The abnormal behaviors in tourism scenes include uncivilized behaviors and dangerous behaviors.Uncivilized behaviors include climbing,doodling,littering and other behaviors.Dangerous behaviors include fence climbing,falling into water and other behaviors.This thesis uses image processing and deep learning to detect and identify the tourist behavior of the monitoring scene in the scenic spot.If there is any abnormal behavior of tourists,the thesis initiates early warning,which provides a good supervision technology for the big data tourism supervision system.Firstly,this thesis proposes a deep learning method based on three-dimensional region convolution network(R-C3D)is applied to the recognition of climbing and drawing in tourism scenes.This thesis describes the climbing and painting behavior,and then introduces the specific steps from two aspects: video preprocessing and behavior recognition model based on 3D convolution network.In the R-C3 D model,it can detect any length of video.At the same time,it can use the characteristics of time sequence generation and network classification to share C3 D convolution features,and achieve the detection speed 3 times faster than the current common methods.The experiments show that,compared with other methods,the deep learning method based on R-C3 D achieves high real-time and high accuracy in climbing and drawing behavior recognition of tourism scene.Then,aiming at the problem that the general behavior recognition methods can not recognize the abnormal behavior in the tourism scene,this thesis applies the behavior recognition method based on Yolo,GOTURN and trajectory analysis process.Firstly,this thesis describes the problem of fence climbing behavior,and then describes it in detail from four aspects: video processing,object detection model implementation,object tracking model implementation and trajectory analysis method implementation.In the Yolo model,after the routine process of One-stage,a single convolution network is used to realize end-to-end human recognition and bounding box position prediction,and the effect of high accuracy is achieved.At the same time,in the GOTURN model,the advantage of off-line learning neural network is used to directly regress the bounding box of the object human body,and the model has better tracking effect for specific categories,which is suitable for human tracking.The experiments show that,compared with other similar methods,the behavior recognition method based on Yolo,GOTURN and trajectory analysis process can improve the detection speed and achieve high real-time performance with high accuracy.Finally,this thesis proposes a real-time multi-human abnormal behavior identification system for the application of real-time multi-human abnormal behavior.The system is based on climbing,painting behavior recognition model and crossing behavior recognition model,and realizes abnormal behavior recognition system.Firstly,this thesis gives an overview of the system,then puts forward the specific implementation of real-time multi person abnormal behavior recognition system,and then describes the implementation method and process of multi person situation.The target loss scheme is given,and the performance before and after improvement is evaluated.At last,this thesis applies the real-time multi person abnormal behavior to the Big Data Smart Tourism Supervision System.It shows the effect of the system applied to the interface,and tests this system.The results show that the real-time multi person abnormal behavior recognition system can be normally applied to the Big Data Smart Tourism Supervision System,and it can effectively warn the abnormal behavior in real time. |
