Design And Implementation Of Real-Time Field Robot Visual Navigation Algorithm Architecute Based On FPGA

The development of the agriculture-based information technology is of necessary technical method to enhance the productivity and smart of the agricultural.Because of the low-cost and rich-information,vision-based navigation algorithm has received a wide attention.Although the vision-based robot navigation method can control the robot to walk according to the strategy,the complex and changeable field environment increased the difficult to detection the crop row accuracy.Moreover,the existing method is too complex to meet the real-time requirements without the use of super computers.Therefore,the purpose of this thesis is to design and implement a navigation algorithm for field robots with low power consumption and fast running speed.The key to realization of vision-based field robot navigation algorithms is quickly and accurately detecting the position of the rows in complex field images.The traditional crop row detection methods are off-line and non-real-time.Therefore,this thesis proposes an optimized method based on sector scanning to detect the position of crop row line in the image.According to the experimental results,the proposed method can effectively detect the position of crop row,and the average accuracy is up to 89.7%.In order to reduce the time consumption,a FPGA-based hardware acceleration for navigation algorithms is designed and implemented.The overall method consists of three parts: image pre-processing,optimized crop row detection and robot navigation parameter extraction.All of the three parts uses parallel pipeline architecture.Among them,the pre-processing module calculates the greenness and threshold in parallel and segments crop from background in pipeline way;while the crop row detection module calculates the row density and the boundary of crop row in parallel,then transfer image data in pipeline.Experimental results shows that the image pre-processing module,which consumes 16 ms,can effectively segment the green crop from background at the resolution of 1920×1080 with F-score is 91.1%.The pre-processing module occupies 3042 triggers,6605 lookup tables,and 4kb block registers.While the crop row detection module can reach the accuracy of 89.7% and only consumes 0.2s.And the navigation parameter extraction module occupies 372 triggers and 1013 lookup tables.A low-cost FPGA-based filed robot navigation algorithm which implemented in this thesis can meet the requirements of real-time and accuracy.