Research On Accelerated Image Processing Technology Of Plant Protection UAV Based On Structural Light Vision

Our country is a big agricultural country.In recent years,with the acceleration of the population aging and the reduction of rural labor force in our country,the traditional artificial plant protection method can not adapt to the increasing demand of agricultural production.Therefore,UAV technology has been applied in the field of agricultural plant protection,and research about this is of great importance.However,typical obstacles such as telephone poles and trees,are often found in the operation scenes of plant protection UAVs,has affected the flight UAV near-field operations.In order to ensure the safety of the UAV,it is required that the plant protection UAV must has the automatic obstacle avoidance capability.Therefore,an obstacle detection method based on structured light vision was proposed.In order to adapt to the speed of on-line detection of plant protection UAVs,a GPU-based research on structured light image of obstacle acceleration processing technology was proposed.The obstacle detection based on structured light vision mainly collects the structured light image of the obstacle through the CCD and performs a series of image acceleration processing through the GPU to obtain the complete single-pixel obstacle profile curve.Finally,the parameters such as the distance,the width and the azimuth of the obstacle were obtained through the parameter calculation so that the plant protection UAV can avoid obstacle according to these parameters.The main contents of this paper are summarized as follows:(1)The existing mainstream UAV obstacle detection technology and image acceleration processing technology are studied,and their respective advantages and disadvantages are summarized.Based on this,the obstacle detection method based on structured light vision is proposed.And this paper focus on the CUDA architecture obstacle structured light image processing methods to ensure the real-time of plant protection UAVs.(2)For the real-time detection of plant protection UAV needs,the embedded GPU is applied to the obstacle image parallel processing.This paper has in-depth study of the CUDA programming model,and complete the obstacle detection system software and hardware design for the parallel realization of image processing laid the foundation.(3)For the enhancement algorithm in the image processing flow,a space-based image enhancement algorithm based on point processing under CUDA parallel computing architecture is proposed.The enhancement algorithm effectively increases the gray level difference between the background and the obstacle.It improves the gray level of the obstacle contour line,at the same time it weakens the influence of the background noise while.Finally,paper analyzes the parallel algorithm performance,and optimizes the parallel algorithm by modifying the implementation of configuration.(4)For image segmentation in image processing flow,a structured light image parallel segmentation algorithm based on grid-based two-dimensional entropy threshold segmentation and k-means clustering algorithm is proposed in CUDA parallel computing architecture.The algorithm takes full advantage of obstacle structured light image spatial information.It effectively ensure the image segmentation effect,and segmentation time can well meet the needs of plant protection applications.Finally,the performance of parallel segmentation algorithm is analyzed,and the parallel algorithm is optimized by modifying the execution configuration.(5)For image thinning in image processing flow,an improved FPA thinning algorithm based on meshing is proposed by comparing with the existing FPA thinning algorithm and according to the characteristics of structured light image,which is very suitable for the light image of obstacle structure.The algorithm is implemented in parallel under CUDA architecture.Finally,the performance of parallel thinning algorithm is analyzed,and the parallel algorithm is optimized by modifying the execution configuration.(6)Aiming at the binary image after the image processing,the method to calculate the distance,width and azimuth detection parameters of the obstacle based on the ROI region is proposed.Firstly,the ROI region of the obstacle is obtained by accumulating the horizontal gray-scale first,and then the coordinates of the end-point of the obstacle contour are extracted from the ROI region to calculate the distance,width and azimuth of the obstacle.(7)Through four aspects of image processing effect,processing time,obstacle detection static experiment and dynamic experiment,the experimental results show that the detection error of obstacle parameters meets the accuracy requirements of the detection and detection system of this paper.The single detection time is about 55 ms,which meets the real-time requirements of obstacle detection system of plant protection UAVs.