Research And Implementation Of Dynamic Gesture Recognition System Based On FPGA

Artificial intelligence has become the hottest topic nowadays,and gesture recognition is one of the important areas of artificial intelligence.The disadvantages of bulky and expensive wearable data gloves limit the usage scenarios,and a large number of gesture recognitions based on machine vision came into being.In gesture recognition,gesture status is divided into static and dynamic,and dynamic gesture recognition is the current research trend.By analyzing the current status of research on dynamic gesture recognition technology at home and abroad,it is found that the speed and accuracy of dynamic gesture recognition is still a big challenge.In order to improve the speed and accuracy of gesture recognition,it is proposed to use FPGA's parallel processing capabilities and field programmable features to accelerate the front-end processing part of dynamic gesture recognition,and the back-end recognition part is handed over to the PC host computer that is good at complex algorithms.The idea of realization.The system first uses the skin colors model and calculates the gesture centroid direction angles for segment the gesture area and extract the gesture centroid feature.By comparing the hidden Markov and dynamic time warping algorithms,the dynamic time warping algorithm is selected as the gesture recognition model.The front end uses FPGA to accelerate,and the back end limits the DTW search range to improve the algorithm speed.Thesis research work mainly includes the following aspects:First,study the current situation and latest achievements of a large number of dynamic gesture recognitions at home and abroad.According to the real-time requirements of the system,a system solution using FPGA as the image processing hardware acceleration platform and PC host computer as the gesture recognition algorithm processing platform is proposed.Second,through the research and analysis of image processing algorithms,combined with the simulation platform,the feasibility of the dynamic gesture recognition algorithm was verified and analyzed,and a simulation model was built for the system.Respectively verify the skin color detection,mathematical morphology processing,Sobel edge detection,centroid feature extraction,and recognition algorithms.Third,take a FPGA chip as hardware acceleration platform,adopt UP TO DOWN design method,Verilog HDL language completes the function design of hardware acceleration part.Including skin color detection module,median filters module,corrosion expansion module,Sobel edge detection module and gesture feature extraction module.During the early simulation and debugging,the image data is sent to the FPGA through the serial port.After being processed by each module,it is finally scanned and displayed on the VGA screen.Fourth,after the FPGA extracts the dynamic gesture feature sequence,the feature sequence is sent to the host computer through the serial port for gesture recognition.Gesture recognition adopts the dynamic regularization model algorithm,and improves and optimizes the algorithm to improve the search efficiency of the recognition algorithm.And combined with Python language to complete the training and recognition of the algorithm.And wrote the system display interface.Fifth,the system is set to train 8 kinds of dynamic gestures,4 kinds of gesture trajectories are linear,and 4 kinds of gesture trajectories are curvilinear.For these 8 kinds of gestures,the test was conducted in 4 different scenarios,and it was concluded that the average gesture recognition rate of the straight line trajectory is about 95%,and the average gesture recognition rate of the curve type is more than 80%.And to evaluate the performance of the system with three indicators: recall rate,precision rate,and precision rate,after the final test,it was found that the system can meet the expected functional requirements and design goals in terms of recognition accuracy and realtime.In this paper,the research achievements of completing dynamic gesture recognition of the help of FPGA have certain significance and reference value for intelligent control scenarios with real-time and contactless requirements.