摘要
运动目标检测中高斯混合模型计算量大、检测实时性较差。为此,提出一种基于匹配次数的运动目标检测算法。利用高斯混合模型构建背景,统计每个像素的观测值与背景模型的匹配次数,根据匹配次数将检测场景分为静态区和动态区,对静态区的像素点进行隔帧检测,对动态区的像素点实行逐帧检测,并结合检测质量和实时性要求研究匹配次数阈值和间隔帧数。实验结果表明,当静态区面积占整帧面积的50%左右时,该算法室内和室外场景每帧图像的检测时间分别为27 ms和20 ms,检测效率较高。
The disadvantage of Gaussian Mixture Model(GMM) is that it requires large amount of computation and its poor real-time detection performance in the process of moving object detection. Aiming at this problem, a moving object detection algorithm based on match times is proposed. It constructs background by creating Gaussian models, counts the number of match times of the observed values of each pixel with the background model, according to the number of match times, divides the detected scene into two areas, such as a static area and a dynamic area. It detects every discontinuous flame for static pixels, does every frame detection for dynamic pixels, and is combined with testing quality and the real-time requirements to study the match threshold value and the interval between frames. Experimental results show that when the static zone area is about 50% for the area of the whole frame, the image detection times of this algorithm for each flame in indoor and outdoor scene are 27 ms and 20 ms, and it can improve the efficiency of detection.
出处
《计算机工程》
CAS
CSCD
2013年第5期192-195,199,共5页
Computer Engineering
基金
国家自然科学基金资助项目(10926719)
河北省科技厅重点实施科技支撑计划基金资助项目(10243554D)
关键词
高斯混合模型
目标检测
匹配次数
实时检测
智能视频监控
Gaussian Mixture Model(GMM)
object detection
match time
real-time detection
intelligent video surveillance