帧间差分与码书模型相融合的运动目标检测算法

Detection Algorithm Based on Consecutive FramesSubtraction and Codebook Model

针对视频检测对实时性和准确性的要求,提出了将帧间差分与码书模型相融合的运动目标检测算法。首先,选取某个t时间内采集到的视频图像作为训练图像,对图像的每一个像素点建立码书模型。把码书模型中表示前景的码书去除,余下的作为背景模型用于检测。检测运动目标时,先将待检测的相邻帧图像进行帧间差分,得到变化区域和没有变化的区域,将有变化的区域与背景模型进行拟合,区分出前一帧运动区域和目标运动区域。更新背景模型时以不同的更新方法对前一帧运动区域和目标运动区域进行更新。

Aimed at the real-time and veracity of current algorithm for detecting moving objects,an algorithm based on two consecutive frames subtraction and codebook model is presented,First,select the video images which are collected in a consecutive time as the training images,establish a codebook for every pixel of the images.From the first two images on,subtract two consecutive frames to find out moving areas and background areas.Compare the moving areas with the background model to decide whether the moving area is the moving objects or not.The background updating algorithm varies in the rate of updating according to the different areas of the image.