Research On Low-speed Unmanned Vehicle Navigation Based On Detection Of Driving Area

With the rapid development of the mobile Internet,people's demand for takeaway and express delivery services is becoming stronger.The number of takeaways and express delivery is increasing.The demand for delivery staff and couriers is also increasing.The low-speed unmanned vehicle can complete the efficient delivery of takeout and express delivery,replace the takeaway and courier to complete the "last mile" delivery task,thereby effectively saving delivery costs and improving delivery efficiency.When unmanned logistics distribution vehicles deliver in the park environment,they mainly detect obstacles by lidar,and then builds a map to complete navigation.However,due to its own working mechanism and performance limitations,lidar is unable to identify the road category,resulting in unmanned vehicles driving on undriving roads,and unable to effectively detect low-height obstacles such as curbs,which affects driving safety.In order to enable unmanned vehicles to safely navigate in the park environment,this paper designs a navigation system based on drivable area detection.The system is mainly divided into two modules,namely the drivable area detection module and the path planning module.The drivable area detection module mainly performs pixel-level segmentation processing on the image collected by the camera,and detects the drivable area in the image.It is easier to detect and identify road categories and low-height obstacles in the image,so the drivable area segmented from the image can effectively avoid nondrivable roads and low-height obstacles.Then,we can projecte the laser point cloud into the image by using the relative pose relationship between the camera and the lidar,extracte the point cloud of the drivable area and constructe the current frame cost map through the path planning module.We update the global static map for global path planning and the local cost map for local path planning by designing a cost map generation strategy,and then use the A* algorithm and TEB(Time Elastic Band)algorithm in the new global static map and local cost map respectively to complete global path planning and local path planning.So that unmanned vehicles can safely avoid obstacles.In order to verify the effectiveness of the navigation system based on the driving area detection,firstly,the driving area detection module of the image was verified,and the test was carried out on the campus scene data set.The experimental results show that the network can extract drivable area accurately in the campus environment,then extract the drivable area point cloud to obtain the depth information of the drivable area.Then,the path planning module of the system was verified and tested in a simulation environment.The experiment showes that,comparing with using lidar alone for navigation,fusion image and point cloud information navigation system based on the drivable area detection by can identify drivable roads and avoid low-altitude obstacles effectively to complete navigation successfully.