摘要
研究了在静止的单摄像机条件下滞留与偷窃物体检测与分类算法。基于轮廓的判别方法在环境轮廓复杂情况下检测率会降低。在吸收了原有轮廓空间相似性算法的基础上,加入了轮廓的连通性判断,只有轮廓同时满足空间和连通性都相似的物体才被判定为滞留物体。此外还研究了基于颜色直方图的巴氏距离的判定方法,将以上两种方法进行了比较。实验结果表明,在现实环境下,改进后的轮廓判别方法比颜色方法适应性更强,检测正确率更高。
Algorithm for real time abandoned/stolen objects detection by a single and static camera was studied. It intended to solve the problem that the detection validity of gradient-based method will decrease under the condition that the gradient of environment is complex. Besides the edge points' position analysis of contour, the edge points' connectivity analysis of contour was introduced. 0nly the object of which the contour reaches the similarity of both the edge points' position and connectivity can be recognized as the abandoned object, tn addition, Histogram-based method which calculates and compares the Bhattacharya distance between histograms to determine if the object is abandoned or stolen was studied, and was compared with the first method. Experimental results show that the improved gradient-based method is more adaptable and effective.
出处
《计算机应用》
CSCD
北大核心
2007年第10期2591-2594,共4页
journal of Computer Applications
关键词
运动目标检测
行为理解
滞留和偷窃物
moving object detection
activity recognition
abandoned and stolen objects