基于伪球算子边缘模型的猪前景帧检测

针对固定区域监控非刚体前景目标的活动,提出一种基于边缘像素模型的、低计算代价的前景帧检测法。根据帧序列中前景帧的边缘像素稳定性远远小于背景像素的特性,首先采用伪球边缘检测算子提取每帧的帧边缘;接着根据帧序列滑动窗口帧数和背景边缘像素判定阈值,提取非背景边缘图像;然后根据边缘像素点数量阈值,消除噪声边缘;最后根据序列帧噪声边缘判断阈值,判断当前帧是否为前景帧。在具有相当平滑性的条件下,伪球算子的边缘定位误差小于canny算子。猪舍区域检测猪只目标的试验表明,采用不到原图像像素总数的3%的边缘像素点,可以有效地适应背景光线变化及前景运动缓慢或短暂滞留等情况,为后继的猪只运动分析创造有利条件。

Detection of foreground-frame of pig using edge model based on pseudosphere-operator

Based on edge model, the method of foreground-frame detection with low computational cost was proposed for monitoring the activities of the non-rigid foreground objects in a fixed region. According to the fact that the pixel stability of the foreground edge was far less than that of the background, the steps of the method was as follows: firstly the pseudosphere-based edge detector was used to extract edges in each frame, and the non-background edges were extracted according to the sliding window frames of video sequence (T) and the threshold of judging background edge pixel (η); then the noise edges were eliminated in terms of the threshold of edge pixels; finally the foreground frame was judged according to the threshold of noise edge pixels. Compared with the canny edge detector, in the case of having the same smoothness, pseudosphere-based edge detector offers a lesser error for edge locating. The experimental results on detecting pig in a pigpen showed that the proposed algorithm could effectively adapt to the background in illumination variation and the foreground in slow motion or short stay. This research provides a reference for the subsequence behavior analysis of pigs.