直方图反向投影多目标检测优化算法  

Optimized Multi-Object Detection Method Based on Histogram Back-Projection

作　　者：张辰[1,2] 赵红颖[2] 钱旭[1] 

机构地区：[1]中国矿业大学(北京)机电与信息工程学院,北京100083 [2]北京大学遥感与地理信息系统研究所,北京100871

出　　处：《计算机系统应用》2014年第2期178-182,218,共6页Computer Systems & Applications

摘　　要：针对目标检测中图像背景信息复杂以及反向投影方法目标颜色类型单一的问题,提出了一种改进的直方图反向投影目标检测优化算法.在建立多个目标模型弥补一般反向投影方法不能同时检测含有不同颜色信息目标这一不足的同时,通过对目标概率图像进行一系列优化处理以减小复杂背景对目标检测的干扰.实验结果表明,该算法在满足对视频图像实时处理要求的同时可准确检测目标位置,且检测效果及计算效率优于传统反向投影算法及其他方法.Aiming at the complexity of background image in object detection and the singleness of object＇s color type in common back-projection method, an optimized object detection method based on histogram back-projection was presented. In this algorithm, multiple object models were established to make up the defect that multiple objects with different colors cannot be detected by using common back-projection method. Besides, in order to reduce the interference of complex background, a series of optimizing process steps for the probability image was added. Experimental results show that the proposed algorithm not only can accurately detect the position of objects in real-time, but also with better detection effect and higher operation efficiency than the common back-projection algorithm and other methods.

关 键 词：目标检测 反向投影 目标概率 颜色直方图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]