摘要
针对当前目标检测算法主要集中在GPU上运行且对家庭机器人任务针对性不强的问题,提出了一种基于OpenCV的RMFC算法。该算法首先对输入图像进行裁剪、腐蚀、Canny边缘检测操作;其次使用二维顺序查找的方法过滤地面边缘以外的区域;随后使用卷积判断模块抽取物体的位置,最后在视频测试集上测试,结果表明该方法在快速提高检测速度的同时兼顾了准确性及针对性,并得到每幅图像少于10个的候选区域。
In order to solve the problem that the current target detection algorithms mainly run on GPU and are not targeted to home robot tasks,a RMFC algorithm based on OpenCV is proposed.Firstly,the algorithm clips,erodes and conducts Canny edge detection to the input image.Secondly,the two-dimensional sequential search method is used to filter the areas outside the ground edge.Then,the algorithm utilizes a two-dimensions sequential search to filter the regions out of the bound of the ground.To extract the position of objects,the convolution and judge method would be processed after that.Finally,after the video data test,it is demonstrated that the algorithm can reach the balance among speed,accuracy,and pertinence.Also,the region proposal in every picture is less than 10.
出处
《科技创新与应用》
2020年第17期1-5,10,共6页
Technology Innovation and Application
关键词
目标检测
候选区域
顺序查找
卷积
object detection
region proposal
sequential search
convolution
