Research On Visual Segmentation In Solid Waste Sorting

There is a large amount of construction waste in China and it is extremely urgent to recycle the resources of construction waste.Solid waste sorting is an indispensable technology in the recycling of construction waste as it can segments and recognizes the objects in solid waste images and guides robots for sorting.However,the existing image segmentation algorithms are not suitable for the specific scenario of solid waste sorting.Solid waste objects are easily obstructed by dust and debris,and their visual features are seriously degraded.The traditional methods of segmentation by color and edge information are almost ineffective.The segmentation algorithms that incorporate RGB-D information also cannot meet the technical requirements for solid waste sorting.In view of the above problems,this thesis proposes an efficient RGB-D segmentation algorithm in harsh environment based on the ambiguous regions extraction.And a further optimization algorithm is also proposed.The work and contributions of this thesis are as follows:1.A data set of solid waste is built.The data set includes four common types of solid waste,and there are large differences in the shape and size of objects.Through the depth camera,RGB images and point clouds of solid waste were collected.In order to quantitatively analyze the segmentation results,the data set is also manually labeled as ground truth.2.Using depth information to build the background conveyor belt model,and propose a solid object segmentation algorithm based on the ambiguous regions extraction.This method can not only extract the ambiguous regions that are easy wrong segmented,but also effectively separate the adhesion solid objects.High-precision segmentation results are obtained by reprocessing the ambiguous region.3.An optimization algorithm is also proposed based on full connected conditional random field model.An energy function is proposed for solid waste objects segmentation,and then a fully connected conditional random field is constructed,and the segmentation of solid waste is achieved globally.For reducing time-consuming,multithread is also used to speed up the algorithm.Finally,an experiment is conducted on the solid waste dataset to analyze the algorithm's segmentation accuracy and time-consuming.The proposed algorithm for solid waste image segmentation is significantly better than state-of-art segmentation algorithms.