Research And Implementation Of Target Recognition Algorithm Based On 3D Point Cloud

In recent years,with the continuous development of three-dimensional technology,the target recognition technology based on three-dimensional point cloud has been paid more and more attention.Compared with the two-dimensional image,the three-dimensional data acquisition is less affected by environmental factors,and the three-dimensional point cloud data can better save the target's coordinates,posture and other important information,and can directly reflect the position of the target in the real space.Therefore,this paper will study the target recognition technology based on 3D point cloud.The 3D point cloud data collected in the actual environment usually contains a large number of noise points,which will have a great impact on the subsequent point cloud processing.Therefore,this paper classifies and denoises the noise points in 3D point cloud data.According to the distribution characteristics of noise points,it can be divided into large-scale noise and small-scale noise.In the process of de-noising,statistical filtering combined with local anomaly factor method is used to remove large-scale noise.Then,fast bilateral filtering is used to smooth the small-scale noise.Compared with the traditional bilateral filtering,the fast bilateral filtering can effectively improve the operation efficiency and nearly double the processing time.The experimental results show that the classification denoising can effectively remove the noise and smooth the point cloud data,and can effectively maintain the geometric characteristics of the scanned object.At present,most 3D target recognition algorithms are too dependent on a single feature,and the description of a single feature on the 3D model is limited,and most algorithms do not take into account the overall and local features at the same time.To solve these problems,this paper proposes a 3D object detection method based on feature fusion.Firstly,orb feature points are extracted to describe the local data characteristics of the model;The second step is to extract Canny edge information and select shape context descriptor based on it;In the third step,the above two features are weighted and fused to obtain a more comprehensive feature;Finally,the final detection result is obtained by calculating the feature similarity.Test results show that this algorithm can effectively improve the detection performance.Compared with the traditional method,the algorithm proposed in this paper works well in ??? and P-R curve and other five evaluation indexes have been improved to a certain extent.Finally,the algorithm is applied in the vehicle chassis security inspection system.For vehicle chassis 3D data,firstly,the large-scale noise is removed by statistical filtering and local anomaly factor algorithm,and then the small-scale noise is filtered by fast bilateral rate wave.Experiments show that the combination of the above filtering methods can effectively filter the large-scale and small-scale noise points in vehicle chassis 3D data,and can effectively retain the main point cloud.At the same time,in order to reduce the number of invalid point clouds and the amount of subsequent processing,this paper compares three sampling methods: grid sampling,farthest distance sampling and geometric sampling.The results show that compared with other sampling methods,geometric sampling can better preserve the edge information of objects,and the algorithm is not sensitive to the data density of point clouds,and the number of samples can be controlled.Finally,a more comprehensive feature is obtained by fusing orb feature and shape context feature,and the vehicle chassis safety detection task is realized by template matching.