基于视觉识别的四通道在线粮包计数仪开发

发布时间：2018-02-24 03:36

本文关键词： 计算机视觉粮包计数目标检测目标跟踪运动预测　出处：《浙江理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：对密集型的粮食生产、加工企业而言,粮包储运出库、入库数量统计的准确性关系到企业的生产效益。而在这些企业的生产线上,粮包的自动、精确计量技术仍然是一个难题。传统的光电粮包计数器,根据“光电传感”原理,通过计算包袋遮挡红外线的时间长短进行计数。但是这种计数方式存在这几方面的问题:有飞虫或杂物遮挡传感器时会对包袋计数产生干扰;两包完全重叠时或三个以上包袋粘连时,计数产生误差;功能单一,无法识别粮包破包等情况。计算机视觉技术可以很好地解决以上存在的问题,因此针对包袋的自动计数问题本文提出一种新的思路:基于计算机视觉技术实现运动粮包在线自动计数。即通过视频分析和图像处理,实时在线对运动包袋进行自动计数,解决连包、漏包情况下计数误差大的问题。通过对通道中包袋的运动情况分析,本文提出了一种基于跟踪检测的包袋计数方法。首先,对进入监控区域的包袋目标进行跟踪,当检测到目标越过第一条检测线时,跟踪检测目标连通区域内独立包袋数目。当目标越过第二条检测线时对通道内流经包袋数目进行精确累计。针对包袋的运动特点,本文提出了基于多项式运动插值函数的质心位置预测方法对包袋目标进行跟踪匹配,并采用了多线程编程技术,实现多目标自动跟踪。同时,研究了修正系数?对预测误差的影响,实验结果显示?取0附近的值时,目标跟踪效果最佳。另外,针对所跟踪的目标连通域内包袋粘连计数问题,提出了基于面积统计和连通域分析的盲分割计数方法,该方法可以有效地检测出包袋粘连情况,并计算出连通区域独立包袋数目。最后,对4通道中包袋流动视频序列进行了实验分析。实验结果显示,在摄像头采帧率在30帧/秒,通道的倾斜程度在25°~35°之间,包袋运动速度大约在35~58cm/s,监控区域内包袋粘连不超过4包的情况下,4通道在线包袋计数仪的计数准确率可以达到98.11%。此外,实验结果验证了本文包袋自动计数方法的有效性,在此基础之上,本文设计了4通道粮包在线计数系统在实际场景中的应用解决方案。
[Abstract]:For the intensive grain production and processing enterprises, the accuracy of storage and transportation of grain and storage quantity statistics is related to the production benefit of the enterprises. In the production line of these enterprises, the grain package is automatic. Accurate measurement technology is still a difficult problem. The traditional photoelectric grain package counter, according to the principle of "photoelectric sensing", Counting by calculating the time of envelope blocking infrared ray. But there are some problems in this counting method: when there are flying insects or sundries to block the sensor, the bag count will be interfered; When two packets are completely overlapped or when more than three bags are conglutinated, the counting error is produced; the function is single and the broken grain package can not be identified. The computer vision technology can solve the above problems very well. In this paper, a new method is proposed to realize the automatic counting of moving bags based on computer vision, that is, through video analysis and image processing, the automatic counting of moving bags is carried out online in real time. By analyzing the movement of bags in the channel, a bag counting method based on tracking detection is proposed. Firstly, the bag objects entering the monitoring area are tracked. When the target crosses the first detection line, the number of independent bags in the connected area is tracked. When the target crosses the second detection line, the number of bags flowing through the channel is accurately accumulated. In this paper, the centroid position prediction method based on polynomial motion interpolation function is proposed to track and match the bag target, and the multi-thread programming technique is used to realize the multi-target automatic tracking. At the same time, the correction coefficient is studied. Experimental results show that? When the value near 0 is taken, the target tracking effect is the best. In addition, a blind segmentation counting method based on area statistics and connected domain analysis is proposed to solve the problem of bag adhesion counting in the connected domain. This method can effectively detect the bag adhesion and calculate the number of independent bags in the connected region. Finally, the experimental analysis of the packet flow video sequence in the four channels is carried out. The experimental results show that the frame rate of the camera is 30 frames per second. The inclination of the channel is between 25 掳and 35 掳, the velocity of bag motion is about 35 ~ 58 cm / s, and the counting accuracy of the on-line bag counter can reach 98.11 when the bag adhesion in the monitoring area is not more than 4 packets. The experimental results verify the effectiveness of this method. On the basis of this, a 4-channel on-line counting system for grain package is designed.
【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TB486;TP391.41

【相似文献】