采用随机样本进行运动目标检测的算法研究

针对目标检测中由于背景光线突变等复杂条件所造成的背景无法实时更新等缺陷，提出了一种基于随机样本的目标检测方法。背景的建立依赖一幅图像中的每个像素点的历史采样值或其邻域采样值，初始化采用像素点的八邻域值来实现。背景的更新采用像素级更新和帧级更新相结合的算法。在正常情况下，采用像素级更新；在复杂条件下，采用帧级更新，且对复杂条件进行了较好的阈值认定。该算法改进了背景建模依赖一段时间内相邻帧的统计值，实现了在单帧进行背景建模，实验结果证明，跟混合高斯模型相比，具有较强的抗噪性和较快的响应速度。

Moving object detection algorithm based on random sample model

To solve the defect of unable to updating background image for complex condition as sudden illumination change in video surveillance, a new object detection algorithm based on a random sampling background model was presented. The background building depended on the sampling of each past pixel and neighbourhood in an image, and its initialization used the 8-neighbourhood, and the updating of the background adopted the combination of pixel and frame updating. Under normal condition pixel, updating algorithm was used, otherwise the frame updating algo- rithm was adopted, and the complex condition was also defined by thresholds. This algorithm improved background building which depended on the statics of consecutive frame in a period and realized building the background in one frame. The experiments indicate that it had better noise immunity and higher speed of response than Gaussian mix-ture model.