Simulation Of Smart Home System Based On Human Gesture Recognition

In recent years,with the continuous progress of virtual reality technology and Internet of Things technology,the field of smart home has made great progress.Home products become more and more intelligent with the improvement of people’s requirements for home living standards.At present,the human-computer interaction mode of most smart homes adopts voice recognition technology or touch interaction technology.Although such technical schemes are widely used,they are often faced with the disadvantages of easy interference of instructions and physical contact.However,for the deaf mute people or patients in infectious disease hospitals,they are inconvenient for voice interaction and direct contact.Hence,the somatosensory interaction technology are proposed in recent years which can better overcome the shortcomings of traditional interaction technology.Moreover,human posture recognition technology has gradually become a state-of-art research in the field of smart home because of its high recognition accuracy,convenient usage,reliable performance and other advantages.Therefore,this thesis proposes a smart home simulation system based on human posture recognition.This system employs a human key points detection module witch processes the human depth image obtained by Kinect depth sensor.By sending the human key point information into the Unity3 D,we simulate the smart home environment with the user’s behaviors.In the simulation,the user’s behavior is recognized and judged in real time.So,users’ gesture commands can be coded and identified.The main contribution of this dissertation are as follows:(1)Overall design of the system.The system employs depth sensors to collect environmental depth information,and uses Unity3 D to simulate the smart home environment.We coded gesture instructions in a virtual environment through the technology of somatosensory interaction,and accomplished the implementation of action instruction registration mechanism.In this case,users can easily register and issue their gesture commands in this system.(2)High accuracy of human body key point localization.We proposed an optimized Mask RCNN structure which achieves faster computing speed and higher accuracy comparing with original algorithm.We target in three parts for optimize,including the improved initial candidate boxes for the scenario considered in this thesis,the improved non maximum rejection ratio to increase the recognition accuracy,and the improved Ro I pooling layer to reduce the operation time while maintaining the recognition accuracy.By comparing the proposed algorithm with he mainstream recognition algorithm,better performance can be achieved in the experiment.(3)Recognition and coding for human gesture instructions.The virtual home environment and virtual human body are modeled by Unity3 D.One of the contribution of this thesis is to recognize the interactive response of human motion instructions and figure out virtual smart home devices.Users’ gesture instructions are recognized and presented by virtual human body in Uniti3 D.Gesture commands can be recognized for controlling the devices in real time.The experiment results indicate that the proposed system can recognize users’ gesture command with high accuracy in real time.Moreover,the proposed scheme has the advantage of anti-interference,high command recognition,simple command release,etc.It is a meaningful attempt to apply human posture recognition technology in the field of smart home,and provides a reference solution for the application of this technology in other fields.