Three-dimensional point cloud instance segmentation method and system and electronic equipment

Abstract

The invention relates to a three-dimensional point cloud instance segmentation method, a three-dimensional point cloud instance segmentation system and electronic equipment. The method comprises the steps of a, inputting point cloud data into a point cloud instance segmentation model, performing feature extraction on the point cloud data by the segmentation model, and outputting semantic segmentation tags of the point cloud data and a high-dimensional vector of each point; b, predicting the object category of each point, and embedding the points into a high-dimensional vector; and c, after vector embedding is completed, predicting the'seed property 'of each point through a seed point selection network, and selecting a better seed point as a reference point to generate an instance to obtainan instance label of each point. According to the invention, a seed point selection network is added; according to the embodiment of the invention, 'seed 'judgment is carried out on each point in thepoint cloud data, and then a better seed point is selected to generate the proposal, so that better instance segmentation is realized, an obvious acceleration effect is achieved for post-processing of a network model, and the problems of low accuracy and low efficiency of a current point cloud instance segmentation technology are solved.

Description

technical field [0001] The present application belongs to the technical field of three-dimensional point cloud data processing, and in particular relates to a method, system and electronic equipment for segmenting three-dimensional point cloud instances. Background technique [0002] In recent years, due to the rapid development of image segmentation technology and neural network, semantic segmentation technology and instance segmentation technology for 2D images have been quite mature, FCN [Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic segmentation.In Proceedings of the IEEEconference on computer vision and pattern recognition(pp.3431-3440).], SegNet[Badrinarayanan,V.,Kendall,A.,&Cipolla,R.(2017).Segnet:A deep convolutional encoder -decoder architecture for image segmentation. IEEE transactions on pattern analysis and machine intelligence, 39(12), 2481-2495.] and other articles have brought methods based on deep learning models to ...

AI Technical Summary

Problems solved by technology

The current technology mainly focuses on the more basic vector aggregation, and only briefly deals with how to divide the instances after aggregation. Therefore, the accuracy of instance segmentation (important indicators) is not very high, and the classification The speed of the

Images