基于SSD的改进目标精定位检测算法

An Improved Object Precision Positioning Detection Algorithm Based on SSD

目标检测问题是计算机视觉中的热门问题,如何提高目标检测定位精度是检测过程中面临的一个难题。在SSD算法的基础上,通过结合选择性搜索算法,提出了一种提高检测定位精度的方法。该算法首先通过SSD算法框架对图像进行目标初始检测,获得目标粗略位置和目标类别,然后采用一种改进的选择性搜索算法对目标所在区域进行选择性搜索,生成目标边界候选框,最后采用文中提出的边界判断方法得到目标精确位置,完成由粗到精(Coarse-to-Fine)的目标定位检测。文中算法对PASCAL VOC2012数据集中的22 531张图像进行了测试,实验结果显示文中算法检测目标中心位置误差7.6,平均覆盖率90.8%,相比于其他算法,中心位置误差更低,覆盖率更高,能提高目标检测定位精度20%～30%。

Detection is one of the important tasks in image processing.How to improve the accuracy of object location is a difficult problem in the process of detection.On the basis of the SSD algorithm,this paper proposes a rough location and object class combined with selective search algorithm.Then,the paper searches the area of object by using an improved selective search algorithm to generate an object candidate frame.Finally,the paper adopts a boundary judgment method to obtain the accurate position of object and complete the detection from coarse to fine.In this paper,a lot of experiments are carried out based on22531 images of PASCAL VOC data sets.The results show that the algorithm has a location error of 7.6 centers and a mean coverage rate of 90.8%.Compared with the other algorithms,this algorithm has lower center location error and higher coverage rate,improving the location accuracy of object detection by 20%~30%.