Unmanned Aerial Vehicle Path Planning Based On 3D Point Cloud

Along with the upsurge of artificial intelligence,the autonomous flight of unmanned aerial vehicles has been widely attention and research,path planning is one of the key links to realize unmanned aerial vehicles autonomous flight.The main task of path planning is to plan the path that meets the constraints of the UAVs based on the mission objectives,the actual flying environment of the UAVs is high-dimensional space,considering the time constraints,the dimension rises to four dimensions,which cause the ”dimension disaster”problem in the planning process.Since the sampling-based path planning algorithm avoids the accurate description of the surrounding flight environment,and the complexity is not affected by the space dimension,this paper uses Sampling-based path planning method for the study of path planning problem of UAVs in dynamic environment.The main contents of this thesis include:1.For image processing of visual navigation influenced by light,this paper establishes a spatial model based on three-dimensional point clouds and analyzes advantages and disadvantages of existing spatial models,such as grid map,geometric map,topological map,and point clouds map.In order to reduce the computational burden of collision detection,the 3D point cloud is converted to an octree-based Octomap for representation of planning space model,and then the collision detection using the bounding box method is performed.2.Based on the establishment of a spatial model,the rapidly exploring random tree algorithm based on sampling is performed in 3D space.In order to solve the randomness of the algorithm,a random probability is introduced to expand random tree with a certain probability of target bias.On this basis,in order to achieve rapid re-planning when encountering dynamic obstacles and make the final path close to optimal,iterative comparisons obtain a random tree where the optimal path is stored.When threats or dynamic obstacles are encountered,it can be quickly local replanning.Experiments show that the improved RRT algorithm performs better in a dynamic environment and the resulting path is shorter.3.The sampling-based probabilistic roadmap method constructs the roadmap before the path planning,an excellent sampling strategy can evenly distribute the sampling points in the entire planning space and construct a roadmap covering the entire planning space to improve effectiveness of algorithm.In order to improve the efficiency of sampling,the existing sampling strategy was analyzed and compared.Finally,the mixed sampling method is chosen to solve the sampling problem of the PRM algorithm in the narrow channel.The experimental results show that the mixed sampling method not only solves the problem of narrow channel,but also gets uniform distribution of sampling points in other free region.Finally,the A*algorithm is used to perform path enquiries.4.The path obtained by the sampling-based path planning method consists of a series of sampling points.In order to satisfy the constraints of flight trajectory of UAVs,path smoothing is performed.Curve smoothing is analyzed and compared aiming at existing Bezier curve and B-spline curve,and then B-spline curve is selected for path smoothing in 3D environment.Finally,through the software design,the entire planning system from drone acquisition to point cloud to final path is realized.At the same time,constraints such as the size of the drone and dynamic obstacles are considered,and the path planning results under different parameters are compared through experiments.