A Machine-Vision Inspection System for Conveying Attitudes of Columnar Objects in Packing Processes

Liang Xu,Xiaomin He,Xiuxi Li,Ming Pan

DOI: https://doi.org/10.1016/j.measurement.2016.02.048

IF: 5.6

2016-01-01

Measurement

Abstract:This paper is a new study on developing a machine vision system for inspecting the conveying attitudes of columnar objects. The presented system consists of image preprocessing, feature extraction, and attitude diagnosis. First of all, in order to segment the objects from the background (namely image pre-processing), an improved maximum between-class variance method is proposed for searching a histogram peak and calculating a threshold value based on the statistics and probability, to solve the problems caused by the non-uniform brightness in a realistic conveyor belt. Then, an open morphological operation is used to eliminate the noise from the binary images produced in the pre-processing step. In the second step (feature extraction), the features of columnar objects are determined by four methods, edge line detecting method, intercepting method, rectangle locating method and feature statistic method. Finally, the diagnosis for the conveying attitudes of columnar objects is based on a hybrid classifier using random forests, and a fuzzy logic. The proposed system is applied to a realistic process for packing industrial explosives. The results of experiments show that the proposed system allows efficient and accurate 100% inspection for the conveying attitude, which ensures the high speed and steady operation of a packing line. (C) 2016 Elsevier Ltd. All rights reserved.