Research Of Object Detection And Behavior Recognition Technology In Intelligent Video Surveillance

Object detection and behavior recognition technology is important research content in intelligent video surveillance.Traditional object detection and behavior technology has been successfully applied in intelligent video surveillance systems.However,in some specific application scenarios,the object detection and behavior recognition technology have the following defects: Primarily,while the traditional object detection is applied to sewage detection,it can only detect whether there exists sewage area in the image and location of sewage qualitatively.Not only the shape of the sewage but also the size of the sewage area cannot be obtained.Besides,Traditional behavior recognition technology is human-centered,it mainly focuses on identifying human's own behaviors,unable to identify human-object interaction behaviors.Aiming at the above two problems,this thesis proposes a sewage detection algorithm based on GAN(Generative Adversarial Nets)and a human-object interaction behavior recognition algorithm based on Deep Learning.These two algorithms are applied to the oil drilling intelligent video monitoring system.Specifically,the main innovations and work of this thesis are listed as follows:1.For the problem of current object detection technology applied to sewage detection,this thesis proposes a sewage detection algorithm based on GAN.The generator uses GAN as the basic network to obtain the size and shape of the sewage area in the image.The sewage level classifier realizes the classification of sewage level based on the feature extracted by the generator.The algorithm adopts multi-task learning.Based on the sewage image dataset constructed by the unique labeling method in the thesis,with one training,the model we get has two functions: sewage area detection and sewage level classification.Experiments show that compared with the existing object detection algorithms,the proposed algorithm can not only detect the location of sewage area,but also obtain the shape and size of the sewage area and identify the sewage level.2.Aiming at the problem that the current behavior recognition algorithm cannot recognize the human-object interaction behavior.This thesis proposes a human-object interaction behavior recognition algorithm on Deep Learning.Based on the combine of object detection algorithm YOLO and 2D person keypoint detection algorithm openpose,we constructed a knowledge base that can be used to recognize human-object interaction behaviors in this thesis.The knowledge base can recognize three kinds of human-object interaction behaviors,including wearing a certain kind of shoes or a certain kind of hat correctly and holding a certain kind of goods in hand.Experiments indicate that the accuracy of proposed algorithm can reach 90%,with excellent availability and reliability.Therefore,it can meet the requirements of intelligent video surveillance applications.3.Based on the two algorithms proposed in this thesis,we did a research about how to apply these two algorithms to the oil drilling intelligent video monitoring system.The thesis mainly focuses on sewage detection and sewage level warning at the drilling site,as well as recognizing whether petroleum worker wear safety helmet and work boots in the work area in order to ensure production safety.The detection result is recorded in the log.When an abnormality occurs,the system can send a real-time warning and store the alarm picture for subsequent investigation and analysis.