一种基于块的校正码书模型

Block based correction codebook model

针对运动物体检测研究的难点之一——复杂动态背景,提出一种新的基于块的校正码书模型。该模型利用HSV空间基于像素块建立校正码书,它具有四个方面的特色:a)引入HSV颜色空间提高了前后景的区分度;b)利用像素块构造码书以克服动态背景对单个像素的影响;c)引入反馈校正机制实现自适应的码书更新,减小伪目标的生成;d)实施码书的小样本学习方法,以提高检测速度。提出测量检测效率的覆盖率—正确率曲线定性评价方法。包含该评价方法的定性和定量实验表明,本模型可以高效快速地检测出复杂动态背景下的运动物体。

Complex dynamic background is very difficult for current moving object detection algorithms.This paper proposed a new block-based correction codebook model to solve this problem,where the correction codebooks were created based on pixel blocks in HSV color space.The merits of the new model lie in four aspects：a）introduced HSV color space to better distinguish the foreground and background;b） used the pixel block to build the codebook,and thus improved the detection performance with the affection of the variation of each single pixel;c）presented a novel correction mechanism so that eliminated false targets efficiently;d）also proposed codebook learning with small samples for fast detection.This paper further proposed a new performance evaluation method called recall-precision curve.The qualitative and quantitative experiments includes this evaluation method demonstrate that the proposed model can efficiently and quickly detect the moving objects under complex dynamic background.