Application of automated image analysis to the identification and extraction of recyclable plastic bottles
Application of automated image analysis to the identification and extraction of recyclable plastic bottles
复制标题
DOI：
10.1631/jzus.a0820788
复制
发表时间：
2009-06-01
影响因子：
3.2
通讯作者：
Mustafa, Mohd Marzuki
中科院分区：
文献类型：
--
作者：
Scavino, Edgar;Wahab, Dzuraidah Abdul;Mustafa, Mohd Marzuki
一种实验机器视觉装置被用来识别和从传送带中提取可回收的塑料瓶。彩色图像是用商用网络摄像头拍摄的，识别是由我们自制的软件根据对象图像的形状和尺寸进行的。该软件能够在一张图片中管理多个瓶子，并进一步扩展到涉及触摸瓶子的案件。识别是通过将测量的特征集与现有数据库进行比较，同时集成各种识别技术，如特征空间的最小距离、自组织映射和神经网络来完成的。该识别系统在一组50个不同的瓶子上进行了测试，到目前为止，对瓶子的识别准确率约为97%。瓶子的取出是通过气动手臂进行的，根据塑料类型激活气动手臂；聚对苯二甲酸乙二醇酯(PET)瓶留在传送带上，而非PET瓶被拔出。考虑到该系统在现有回收厂的商业化，该软件旨在为实时应用提供可靠性和速度之间的最佳折衷。
An experimental machine vision apparatus was used to identify and extract recyclable plastic bottles out of a conveyor belt. Color images were taken with a commercially available Webcam, and the recognition was performed by our homemade software, based on the shape and dimensions of object images. The software was able to manage multiple bottles in a single image and was additionally extended to cases involving touching bottles. The identification was fulfilled by comparing the set of measured features with an existing database and meanwhile integrating various recognition techniques such as minimum distance in the feature space, self-organized maps, and neural networks. The recognition system was tested on a set of 50 different bottles and provided so far an accuracy of about 97% on bottle identification. The extraction of the bottles was performed by means of a pneumatic arm, which was activated according to the plastic type; polyethylene-terephthalate (PET) bottles were left on the conveyor belt, while non-PET bottles were extracted. The software was designed to provide the best compromise between reliability and speed for real-time applications in view of the commercialization of the system at existing recycling plants.