Research And Implementation Of Gesture Recognition System Based On FPGA

With the diversification of image processing algorithms and the mature development of computer processing technology,human-computer interaction technology has developed rapidly,from the beginning of a single mouse control to today's diversified interaction mode.Visual-based interaction is a hot research direction in the field of human-computer interaction,including gesture recognition and behavior recognition.Gesture is a part of human body language.People can get the corresponding meaning by defining gestures.For machines,gestures are simpler and easier to learn than words.As an intuitive and natural way of communication,gestures are favored by researchers in the field of human-computer interaction.As a hardware microprocessor,FPGA is more and more used in image processing and the realization of neural network.In this thesis,we use FPGA to implement the algorithm designed in gesture recognition,and complete the recognition function of gesture,which greatly improves the speed and efficiency of recognition.Firstly,according to the steps involved in gesture recognition system,this thesis introduces and analyses the corresponding algorithms,including gesture image preprocessing algorithm,gesture segmentation algorithm,feature extraction and recognition algorithm,and compares the characteristics of each algorithm.In this thesis,ten kinds of digital gestures are defined in advance.Aiming at the gesture studied,a suitable algorithm is selected.On this basis,a gesture recognition algorithm based on skin color model and BP neural network is proposed.Based on the design of FPGA,the hardware recognition system of gesture is constructed.The hardware platform chooses the black gold development board AX515,collects gesture images through the OV5650 camera,LCD display screen real-time display of the gesture processed by the system,and finally display the recognition results through the digital tube.According to the recognition process,the system is divided into five modules:image acquisition module,preprocessing module,gesture segmentation module,feature extraction module and gesture recognition module.Secondly,according to the design characteristics of the FPGA,the image acquisition and display module,image preprocessing median filtering algorithm,skin color segmentation algorithm,morphological operation,Canny edge detection,Hu moment invariant feature extraction algorithm and BP neural network are designed,simulated and realized.Among them,the complex operations involved in Canny edge detection and BP neural network are simplified and completed by pipeline.Because of its relatively large and complex computational load,the hardware implementation of Canny edge detection and BP neural network takes a lot of time.Finally,after building the sub-modules needed by the hardware system,the system tests and verifies ten gestures by means of software and hardware respectively.On the one hand,the processing effect of the algorithm and the recognition rate of the software are obtained.On the other hand,the recognition results of the hardware and the consumption of device resources are obtained.Through the test results,the recognition rate of the hardware of the system is obtained.It has been proved that the system achieves the basic functions of the design and has obvious advantages over software in speed.