Development Of Four-channel On-line Grain Package Counter Based On Vision Recognition

For the intensive grain producing and processing enterprises,the accuracy of the statistics on the amount of the grain packages' storage and delivery is related to the business benefits.On these enterprises' production line,automatic and accurate counting technology of grain package is still a problem.Traditional photoelectric grain package counter calculates the number of grain package by measuring the time of the bag to block infrared.However,there are several problems of this way: it can not exclude interference of insects and other debris;when the two packages are completely stacked together or when more than three bags are conglutinated with each other,the counting result may be incorrect;it's function is too simple to detect the broken bags and some other situations.Computer vision technology can solve the above problems.Therefore,in this paper,a new idea is proposed: counting the online moving grain packags by using computer vision technology.It is to count the online moving bags automaticly by video analysis and image processing.By analyzing the movement of the bag in the channel,a method of bag counting based on the tracking and detection is proposed in this paper.Firstly,the moving bag object which comes into the surveillance area will be tracked.When a target pass through the first test line,the tracking system will detect and count the number of bags in the connected region through an area-based strategy followed by a region analysis.When the target pass through the second test line,the number of bags moved through the channel is counted and refined.For the features of the bag's movement,a method of object's centroid location prediction based on polynomial motion interpolation function is proposed to track the moving bag object.Multithreading technology is used to achieve the goal of high efficient tracking of multiple targets.Meanwhile,the influence of the correction coefficient on the prediction bias is considered.The experiments' result shows that when the value of ? is near the zero,the prediction effect of this method is best.Moreover,for the problem of counting the number of bags in the connected region which is tracked,an area-based strategy followed by a region analysis is proposed.This method can effectively detect the bags' adhesions and calculate the number of independent bags in connectedregion.Finally,the result of experiments of four-channel online grain package counting is analyzed in this paper.The results show when the camera frame rate is 30 fps,the inclination of the channel is more than 25° and less than 35° and the bag adhesion is not more than 4 packets,the counting accuracy can reach 98.11%.The speed of packages is about 35~58cm/s.In addition,according to the experimental situation,the design of the solution of 4-channel on-line counting system in the practical application is presented.