基于HOG特征与连通区域分析的工件目标检测算法

Workpiece Target Detection Algorithm Based on HOG Feature and Connected Region Analysis

为了解决当前工件目标检测技术在特征不清晰且背景干扰强度大的情况下导致其定位精度不高的问题,文章提出了基于方向梯度直方图(histograms of oriented gradient,HOG)特征和连通区域分析的工件目标检测算法。首先,对训练集中的标准工件目标图像进行网格划分,计算网格内像素梯度,统计梯度直方图,完成HOG特征提取与训练。然后,对单峰阈值的区间划分进行细化,提出了双阈值分割机制,结合连通区域分析,将二值图像转换为标签图像,并进行像素属性评估,滤除干扰,达到准确定位工件目标的目的。最后,基于开源图像库ITK和底层编程实现算法,并集成于标准化软件系统。实验测试结果显示:与当前主流工件目标检测技术相比,文中算法拥有更高的准确性与稳定性。

In order to solve the problem of lowlocation accuracy induced by unclear feature and large background interference in current workpiece target detection technology,the workpiece target detection algorithm based on HOG feature and connected region analysis was proposed in this paper. First of all,the training focus on the meshing,the standard workpiece target image in the computational grid pixel gradient,gradient histogram statistics,complete the HOG feature extraction and training. Then,based on connected component analysis and the double threshold segmentation,convert the binary image to label the map,and pixel property assessment,filter out interference,achieve the purpose of accurate positioning target artifacts.Finally,based on open source image library ITK and the underlying programming algorithm,and integrated with standardized software system. Test results showthat compared with the current mainstream workpiece target detection technology,this algorithm has higher accuracy and stability.