| With the acceleration of rural urbanization and ageing population,the quantity of labor force engaged in agricultural production is seriously insufficient,so it is urgent for intelligent and precise agricultural equipment technology.Undoubtedly,agricultural robot technology is one of the effective ways to solve this problem.According to unstructured characteristics of country roads in Shaanxi Guanzhong area such as no obvious lane mark,no clear and regular boundaries,irregular shap,various breakages and cracks,uneven colors or texture,and so on,algorithm design and simulation were carried out on unstructured road detection,obstacle detection,moving obstacle tracking,road departure detection,obstacle ranging and navigation obstacle avoidance which involved in agricultural robot autonomous navigation on country roads.It is expected to provide technical supports for autonomous navigation of agricultural robots.The main contents and results are as follows:(1)Segmentation and online detection of unstructured roads were studied.In order to reduce the interference of external environment on road segmentations,an unstructured road segmentation method based on super pixels was proposed,and applied to uniform illumination,uneven illumination,water stains,cracks,shadows and other five kinds of country road image segmentations.The results of simulation test showed that the proposed algorithm could maintain the actual boundary well and reach the average accuracy rate of the five kinds of images by 81.47% under the average segmentation time of each image about498 ms.At the same time,in order to improve the real-time performance of the road detection algorithm,an online detection method using temporal and spatial correlation of road regions between consecutive frames based on fuzzy support vector machine incremental learning was proposed.Samples of violating Karush-Kuhn-Tucker and low membership degree were selected to train.The results of simulation test indicated that the proposed method was more robust to the five mentioned country roads,the average accuracy rate was 84.64% which was1.74% bigger than that of no-incremental training,and the online detection speed was 28fps(2)Obstacle detection and tracking algorithms on country roads were studied.According to unstructured characteristics of agricultural robot moving environment,an obstacle detection algorithm based on intuitionistic fuzzy distance was proposed.Experimental results showed that the proposed algorithm could be better to solve the impact of external environmental factors than Otsu method,the Hamming distance method and the exponentialintuitionistic fuzzy divergence algorithm,the average accuracy rate of the proposed algorithm was biggest(85.73%)under the average segmentation time of each image about 100 ms.At the same time,in order to solve the influence of rotation,deformation,torsion and background noise on moving obstacle tracking,the idea of Ada-Boost for reference,selecting samples and a random tree with a minimum error rate was used to learn weight distribution.And an improved hough forest construction method based on sample weights and random tree weights was proposed and applied to track moving obstacles.The experimental results showed that the proposed algorithm could recognize the moving obstacle tracking on country roads,the accuracy rate was 84.44% which was 4.44% higher than that of the traditional hough forest,the obstacle tracking processing speed was 25 fps.(3)Road departure detection and obstacle ranging were studied.In order to solve the camera calibration process with complicated corresponding points,large error and low precision,and reduce the influence of bumpy camera state of agricultural robot during running,a monocular model based on pinhole imaging was constructed.A road departure detection method based on road slope boundary line was derived,and an obstacle distance measurement method based on geometric relationship between road and image was also deduced.The results of simulation test showed that the departure detection method was simple and easy to use,and it could accurately reflect the deviation of agricultural robot on the road.Additionally,when the actual distance between the agricultural robot and obstacles within 7m,the absolute error between the distance calculated by obstacle distance measurement method and the actual distance was not exceeding 10 cm.(4)Navigation obstacle avoidance algorithm for agriculture robot was studied.A hierarchical multisensor information fusion algorithm was proposed and applied to control the navigation obstacle avoidance to solve the problem that agricultural robot continued to navigate along the guided path after avoiding obstacles effectively.Firstly,the neural network was used to make fusion of the data level information for distances detected by multiple ultrasonic sensors and calculated by the obstacle measuring model respectively,in order to eliminate the uncertainties of sensor data.And then,the fuzzy neural network was employed to make fusion of the decision level information for the calculated road departure information and the fused distance information.The obtained control signal was more suitable for agricultural robot navigation obstacle avoidance system.The results of simulation test showed that the absolute error between the neural network data fusion distance and the actual distance was no more than 7cm when the actual distance between the agricultural robot and the obstacles within 7m in about 100 ms.And the target error quickly converges to 3.33e-02 in300 ms when fuzzy neural network parameters of membership function were estimated after100 times iterative training.In addition,the proposed algorithm could make the agricultural robot track the central line of the road quickly and avoid the obstacles effectively. |
PDF Full Text Request