摘要
在图像目标识别和跟踪任务中,连通域标记算法的设计优化主要体现在执行速度、存储空间和逻辑判断次数三个方面,因此提出并实现了一种基于一次逐像素扫描连通域标记的单目标检测算法。算法结合包围盒和单目标图像检测的特点,只需单行的图像缓存空间,同时简化复杂的等价标号替换操作,目标的判断准则为目标图像连通区域面积最大化,最终以包围盒形式给出目标位置。FPGA仿真结果表明:该方法资源占用率小,检测一幅图像的总时钟周期数为M×N×4(M,N分别为图像行列数),适用于单目标图像的实时识别与跟踪。
In the task of image target recognition and tracking,the design optimization of connected-domain labeling algorithm is only reflected in three aspects：executing speed,storage space and numbers of logical judgment. That＇s why a singletarget detection algorithm based on single per-pixel scanning connected-domain labeling is proposed and realized. The algorithm combines the characteristics of bounding box and single-target image detection,which only requires a single-line image caching space,while simplifying the complex equivalent label replacement operation. The target judging criteria is the maximized area of target image connected region,and the target location is given in the form of bounding box finally. The FPGA simulation results show that the method has less resource consumption rate and total clock cycle number of detecting an image is M×N×4（M and N express the number of image rows and columns respectively）,which is suitable for real-time recognition and tracking for single target image.
出处
《现代电子技术》
北大核心
2015年第20期71-74,共4页
Modern Electronics Technique
关键词
连通域标记
FPGA
目标检测
包围盒
connected-domain labeling
FPGA
object detection
bounding box
