| As one of the largest agricultural countries in the world,China has a considerable amount of agricultural resources.With the transfer of rural labor population,the acceleration of the process of agricultural scale and the rapid development of science and technology,the level of agricultural mechanization has been greatly improved.At present,in the four major agricultural production links of "cultivation,planting,management and harvesting",the degree of mechanization of agricultural production management is relatively low,which is mainly reflected in agricultural plant protection.The traditional manual operation mode has extensive modes and low efficiency.In recent years,the emergence of plant protection UAV has greatly improved the efficiency and quality of plant protection operations.Among them,the many advantages of rotary wing drones make it stand out in the field of plant protection,but from the perspective of application status,most UAV for plant protection is remote controlled manually.Obviously,this method is excessively relied on pilot experience and prone to cause "aircraft bombing" accidents once beyond the field of view.However,the centralized management mode of plant protection UAV based on ground stations has gradually become popular,although it can make plant protection UAV fly according to the pre-planned route,but the unknown obstacles that appear during the flight will still pose a huge threat to the safe flight of the UAV.Therefore,the rotary-wing plant protection UAV must have the ability to avoid obstacles autonomously and return to the original route for the continuous plant protection operations after avoiding obstacles.To this end,this paper studies the autonomous obstacle avoidance technology of multi-rotor plant protection UAV based on the fusion of millimeter-wave radar and binocular vision sensors.The millimeter-wave radar is used for early warning of obstacles in the flight direction.Further,the front obstacle recognition is completed through binocular vision.The recognized three-dimensional environment information is projected to a two-dimensional grid plane to form a local navigation map,and finally an obstacle avoidance path planning algorithm based on a pre-planned route is used to complete obstacle avoidance and return to the original operation route.The mainly work of this paper includes:(1)Combining the characteristics of plant protection UAV,an effective target screening method for millimeter-wave radar is researched.The effective target primary selection is carried out by setting the route warning distance.Furthermore,determine whether the effective target in the previous cycle and the current cycle is the same obstacle through the filtering algorithm and the judgment criterion,so as to make the plant protection UAV always focus on the dangerous targets.Experimental results show that the improved Sage-Husa adaptive filtering algorithm has high dynamic target tracking accuracy,and this method can effectively screen out dangerous obstacles within the warning distance.(2)A method for adjusting the main detection direction of millimeter-wave radar device is researched.According to the current pitch angle,flying height of the plant protection UAV and the vertical detection angle of the millimeter wave radar,adjust the main detection direction of the radar so that the plant protection UAV can effectively detect the obstacles in flight direction,and meanwhile,reduce misjudgment caused by crop clutter,so as to improve the effectiveness of millimeter wave radar obstacle detection,and then improve the plant protection efficiency.(3)A binocular stereo matching algorithm based on cross-scale adaptive guided filtering is researched.The improved ETCensus transform and matching cost calculation method can improve the matching accuracy of images in similar texture area by constructing an isosceles triangle to change the comparison pixels of the traditional Census transform;the improved matching cost aggregation algorithm based on adaptive guided filtering can improve the matching accuracy of images in weak texture area by constructing an adaptive shape cross window;the improved adaptive guided filtering algorithm based on cross scale can further improve the matching accuracy of the images in dense texture,similar texture and weak texture area by calculating the matching cost and aggregating the matching cost on multi-scale;Finally,the best stereo matching effect is obtained by parallax post-processing.(4)A three-dimensional environment representation method based on height reduction is researched.Based on the disparity map obtained by stereo matching,the grid method is used to obtain the initial grid map,and the three-dimensional obstacles are projected to the two-dimensional obstacle grid plane through height reduction,which solves the contradiction problem that the accuracy of the three-dimensional environment description and the efficiency of obstacle avoidance path planning can not been obtained at the same time.Meanwhile,it is consistent with the background that plant protection UAV alwalys flies on a certain height from the crops,which can meet the needs of farmland three-dimensional environment modeling for plant protection operations.(5)A multi-rotor plant protection UAV obstacle avoidance path planning method based on pre-planned routes is researched.The improved artificial potential field obstacle avoidance path planning algorithm takes into account the danger degree of obstacles and the route gravitation to obstacles,which can reduce the avoidance times of plant protection UAV and make it return to the original route after obstacle avoidance.Aiming at the problems existed in the artificial potential field,the autonomous obstacle avoidance algorithm with operation accompaniment is proposed,which can improve the operation quality and efficiency of plant protection UAV through constructing the multi-resolution grid map and adopting the idea of working while avoiding obstacles.(6)The software and hardware platform of the multi-rotor plant protection UAV obstacle avoidance system is designed,the obstacle avoidance strategy of millimeter wave radar and binocular vision fusion is proposed,and the target screening of millimeter wave radar,obstacle recognition based on binocular vision and obstacle avoidance experiments have been carried out,and the results have reached expectations which verify the feasibility and effectiveness of the relevant theories and methods in this article. |
PDF Full Text Request