Research On Key Technology Of Motion Recognition For Smart Home

In recent years,computer vision technology has been playing an increasingly important role in the fields of medical care,human-computer interaction and automated video surveillance.With the widespread application of intelligent video systems in real life,video analysis has also attracted extensive attention.Although researchers at home and abroad have made fruitful achievements in the field of motion recognition,human motion recognition is still one of the most challenging problems in computer vision.The diverse shooting angles,complex human-object interaction,and the changing background environment all increase the difficulty of this subject.In this paper,motion recognition technology is applied to smart home system,aiming to increase the security and comfort of smart home by using computer vision technology.This paper mainly proposes three improvements: 1.Control the startup and shutdown of smart home devices through motion recognition technology.2.2.Use movement recognition technology to observe whether dangerous actions such as falling occur in and out of the family.3.Supervise and correct daily exercise of family members through movement recognition technology.In order to achieve the above functions,this paper mainly researches the following aspects:(1)KINECT based motion recognition algorithm,which can realize the intelligent control function in the smart home.In this paper,KINECT device is used to obtain the information of the key points of the human body,which is mapped into the 3d space,and the spatial position relationship and Euclidean distance of key points are calculated to judge the human body posture.Then,TCP/IP and CAN bus protocols are used to control the movements of smart home devices.(2)Small-range motion detection algorithm,which is used in the smart home security system.The security system designed in this paper USES motion recognition technology to detect dangerous actions in the video.Optical flow feature extraction can obtain the motion information of objects in continuous frames.Optical flow feature is an important descriptor for motion detection.However,a part of the noise may be obtained from the background,so this paper USES the CAM based target detection algorithm to extract the optical flow characteristics of the area where the character is located and eliminate the noise extracted from the background.Finally,the third-stream CNN network is designed to extract time-space information.In this part,Res Net network is used as the basic network.In the system,three CNN sub-networks are designed as 3D-Resnet18 network to extract the motion change information along the timeline.(3)Human-object interaction detection algorithm,which is mainly applied in the intelligent monitoring system of smart home system.When detecting the action of human-object interaction,the small range detection algorithm mentioned above will eliminate the motion information of the moving object,leading to the error of the recognition result.Therefore,this paper increases the method of moving target detection to make up the motion information of moving target.This method is a background detection method based on THE Vi Be algorithm,which can effectively separate the foreground object from the background and obtain the displacement information of the moving object between adjacent frames.Later,this paper designs the spatial information of dual-stream CNN network extraction.It USES Res Net network as the basic network and designs two CNN sub-networks in the system.One of the subnetworks is the 3D-RESnet network,which is used to extract the motion change information along the timeline.The other subnetwork is the 2D-RESnet network,which is used to supplement the 2-d spatial information of the recording.To sum up,this paper uses Kinect-based motion recognition technology to realize motion recognition and control of smart home devices,and USES motion recognition technology based on video analysis to realize intelligent security functions and intelligent monitoring functions of the smart home system.In Kinect-based motion recognition technology,this article USES the spatial position relationship of key points and Euclidean distance to determine human body posture.In the motion recognition technology based on video analysis,this paper fully extracts the motion information contained in the video by using the optical flow detection of the person’s area and the moving target detection method,and then classifies the extracted motion information by using the dual-flow CNN model to complete the motion recognition.