Bottle Body Detection Algorithm And Application In High Speed High Accuracy Empty Bottle Detection Robot

Glass bottle is one of the major packaging methods in China,who is one of the biggest beverage consumption country.Empty bottle quality inspection is required before filling the bottle.Human inspection suffered from low efficiency,poor working environment or missed inspection.By using machine vision rather than human eye,this paper studied bottle body inspection in a part of the empty bottle quality inspection.The paper provides two ways to inspect bottle body:one is using the whole glass body as the detection object,the other is dividing bottle body into seven parts to detect separately.Using the whole glass bottle body as the inspect object,first,positioning bottle body by grayscale projection method whose accuracy and efficiency has been proved by a variety of bottle type.Next,input bottle body parameters to establish bottle body model and extract inspection part.Then the paper use two way to detect the defects:the first one finish the inspection by means of threshold segmentation and edge correction;The second one extract main component of bottle body image through PCA dimensionality reduction,and put the main component as input into BP neural network for classification.Both the methods have achieved bottle body inspection,the first method is fast,but the accuracy is relatively low;the second method is with high accuracy,but the dimensionality cost long time and requires a large sum of samples to support.Multi-part empty bottle body inspection divide bottle body into seven areas according to the grayscale distribution of bottle body,and classified as smooth area,wear area,pattern area and LOGO area.Different detection algorithms are proposed for different parts.For smooth area,the paper proposed a method of eight neighborhood gradient threshold to detect defects;and for the wear zone,similarly,a method of horizontal gradient threshold is proposed.And for the pattern area,according to the phase consistency,the paper applied a phase-only based transformation method to remove the pattern and convert the image into smooth area to inspect.Lastly,for LOGO part,the paper extract the character parts and defective parts by improved Canny operator,and then find HOG features of the two different parts,using the HOG descriptors to train Support Vector Machine with the RBF kernel,and finally achieve the inspection by give the classification.Multi-part inspection is highly efficient and accurate.This paper studied the bottle body inspection algorithm of high-speed and high-precision empty bottle inspector in two ways:use the whole body as the object and use 7 part as objects.The first one is fast but the accuracy is not enough in specific areas;the second one finished multi-part inspection,with a little speed sacrifice,and make up the blank of the domestic characters and lines glass bottle inspection.