Research On Key Technologies Of 3D Point Cloud Filtering And Target Detection In Unstructured Environment

Environment sensing serves as the base for automation of construction vehicles in unstructured environment.As the lack of regular shape,color and other obvious characters,the objective recognition is difficult for practical applications.Besides,the accuracy and efficiency of the recognition are significantly affected by the irregular vibration when the vehicles operated in unstructured environment.Facing the above difficulties,a Lidar-based objective recognition method was established in this paper validated by physical prototype experiments.This research is supported by the NSFC project “Research on key technologies realizing autonomous operation of construction vehicles in unstructured environment”.The contents of this paper can be summarized as the following aspects:(1)The literature review on objective recognition and lidar data processing were taken in the first step.Combining with the characters of unstructured environment,construction vehicles and Lidar techniques,the outline of this research was built up.(2)In order to diminish the noise with high value of discreteness from the original lidar point clouds data effectively and adaptively,a modified multi-layer filter was established with the function of selecting the threshold value of standard deviation autonomously based on the evaluation of how the Gaussian standard deviation influence the data distribution.Additionally,for the purpose of maintaining the character of the lidar point cloud data on the objects,a modified Bilateral filtering algorithm was applied with the value of weighting factor determined by the normal cosine between different regions of the point clouds.(3)Based on the information of the objects characters,a frequent objective recognition method was established.Firstly,the region of the objects was detected by analyzing the information of the edges.Secondly,the height of the objects was calculated through the fitted curve of the objects.Finally,the direction of the objects was determined by the coordinate transformation.Therefore,the information of the objects(length,width,height and direction)can be gained based on the rough rata of lidar in unstructured environment.(4)A testing device was constructed and applied for the validation of the objective recognition method established in this research.Three operating conditions and two driving modes were taken for the tests on the accuracy,efficiency and stability of the method.Tests results indicated that the method can be applied in different conditions with acceptive accuracy.An objective recognition method for the application of construction vehicles in unstructured environment was built in this paper with the performance verified through field tests.This research provides a reference for the development of construction vehicle automation.