How Far are We from Solving Pedestrian Detection?

CVPR 2016 我们离解决行人检测问题到底还有多远?

How Far are We from Solving Pedestrian Detection?
项目网页:https://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal-computing/research/people-detection-pose-estimation-and-tracking/how-far-are-we-from-solving-pedestrian-detection/

Matlab code: https://bitbucket.org/shanshanzhang/code_filteredchannelfeatures

行人检测问题一直是一个比较热门的研究话题,行人检测最近几年进步比较大,那么还有多大的提升空间了?最近建立了一个 human baseline(人自己去看,来框出行人),发现大约目前最先进的算法与人工检测差 10%。

How Far are We from Solving Pedestrian Detection?_第1张图片

于是我们分析了一下目前算法主要存在的问题,找出改进的空间:1) high confidence false positives 一个主要因素是 localisation ,这里我们通过改进 training set alignment quality 来解决;2) 背景的干扰,通过深入分析CNN网络来改进

这里我们首先建立了一个 human baseline, 这是我们的终极目标,达到或者超过人工检测率。

How Far are We from Solving Pedestrian Detection?_第2张图片

How Far are We from Solving Pedestrian Detection?_第3张图片

3.2. Failure analysis

紧接着我们分析了一下 当前最好的算法检测失败的原因
How Far are We from Solving Pedestrian Detection?_第4张图片
How Far are We from Solving Pedestrian Detection?_第5张图片

Conclusion: For most top performing methods localisation and background-vs-foreground errors have equal impact on the detection quality. They are equally important.

3.3. Improved Caltech-USA annotations
原来的数据库真值有不太准确的地方,我们对此进行了改进。

4.1. Impact of training annotations
How Far are We from Solving Pedestrian Detection?_第6张图片

4.2. Convnets for pedestrian detection

How Far are We from Solving Pedestrian Detection?_第7张图片

How Far are We from Solving Pedestrian Detection?_第8张图片

Conclusion :CNN网络在图像分类和广义目标检测问题显示出很强的能力,但是针对小目标检测定位问题表现的不是很好,加入了Bounding box regression (and
NMS)有所改善,但是 背景的干扰仍然是检测失败主要的原因。

你可能感兴趣的:(行人检测,ZJ,CVPR,2016)