编辑: hgtbkwd | 2019-07-13 |
08, CVPR'
09, ICCV'
09, PC of ICCV 2011Research interestPerceptual computing, human-computer interfacesPapers… * Outline AuthorsAbstractMain contributionsAlgorithmsExperimentsConclusion * Abstract (1/2) Despite impressive progress in people detection the performance on challenging datasets like Caltech Pedestrians or TUD-Brussels is still unsatisfactoryIn this work we show that motion features derived from optic flow yield substantial improvements on image sequences, if implemented correctly―even in the case of low-quality video and consequently degraded flow fields Furthermore, we introduce a new feature, self-similarity on color channels, which consistently improves detection performance both for static images and for video sequences, across different datasets.
In combination with HOG, these two features outperform the state-of-the-art by up to 20%. * Abstract (2/2) Finally, we report two insights concerning detector evaluations, which apply to classifier-based object detection in general First, we show that a commonly under-estimated detail of training, the number of bootstrapping rounds, has a drastic influence on the relative (and absolute) performance of different feature/classifier combinationsSecond, we discuss important intricacies of detector evaluation and show that current benchmarking protocols lack crucial details, which can distort evaluations * Outline AuthorsAbstractMain contributionsAlgorithmsExperimentsConclusion * Main contribution First, we introduce a new feature based on self-similarity of low level features, in particular color histograms from different sub-regions within the detector windowThe second main contribution is to establish a standard what pedestrian detection with a global descriptor can achieve at present, including a number of recent advances which we believe should be part of the best practice , but have not yet been included in systematic evaluationsOur third main contribution are two important insights that apply not only to pedestrian detection, but more generally to classifier-based object detection. (1)Bootstrapping is very important. (2)The existing evaluation protocol is insufficient * Outline AuthorsAbstractMain contributionsAlgorithmsExperimentsConclusion * Outline 本文的风格与该实验室文章一贯的风格类似在自己提出的两个数据库上(Caltech Pedestrian, TUD-Brussel)测试当前人体检测领域不同的特征与不同的分类器,评价这些算法的优劣(性能越高的算法关注度越高)自己提出新特征并通过实验给出结论―― 在原始方法的基础上引入我们的特征可以进一步提升人体检测系统的性能 Related FeaturesHaar-like, VJ 2001年成功用于人脸检测领域HOG (Histogram of Oriented Gradient), Dalal 2005年成功用于人体检测领域HOF (Histogram of Flow), Dalal 2006年提出,应用于视频人体检测HOG-LBP 王晓宇 2009年应用于人体检测领域,高性能CSS (Color Self-similarity), 本文提出Related ClassifiersSVMMPLBoost (Multiple Pose Boosting), Dollar 2008年提出 * Haar-like feature (1/2) Haar-like feature图像内部特定模式的两个矩型内部像素和之差采用积分图可以快速计算Haar特征响应值Haar特征的变种45, 22.5, 11.25度…,仍然受限于 矩形 任意多边形区域形状的Haar特征(CVPR10) Haar特征的积分图计算 传统Haar特征 * Haar-like feature (2/2) 任意形状的Haar特征任意多边形区域的像素和可以等价为一系列梯形区域的像素和梯形区域的像素和等价于两个直角三角形的像素差算法关键是计算直角三角形区域的积分图,参数(x,y,斜率) * HOG feature (1/1) HOG feature-梯度方向直方图输入图像的Gamma校正计算输入图像各像素的梯度幅值与方向梯度幅值高斯加权,使用三线形插值计算各个单元梯度方向的直方图相邻的单元直方图归一化得到最终的特征向量 HOG特征计算流程 HOG特征的三线性插值 * HOF feature (1/1) HOF feature-光流直方图计算输入图像的x、y方向的光流 (例如LK算法等等)对于特定区域对,根据对........