How Far are We from Solving Pedestrian Detection?

冷不防 2022-06-03 03:37 146阅读 0赞

原文链接 http://blog.csdn.net/cv\_family\_z/article/details/52119644

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/][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][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%。

![这里写图片描述][20160804135917943]

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。

![这里写图片描述][20160804142412315]

![这里写图片描述][20160804142520861]

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   
![这里写图片描述][20160804142856300]   
![这里写图片描述][20160804142902995]

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   
![这里写图片描述][20160804143758054]

4.2. Convnets for pedestrian detection

![这里写图片描述][20160804151721930]

![这里写图片描述][20160804151747515]

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。

[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]: 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/
[https_bitbucket.org_shanshanzhang_code_filteredchannelfeatures]: https://bitbucket.org/shanshanzhang/code_filteredchannelfeatures
[20160804135917943]: /images/20220603/83740d415cd9442c885463d94b504296.png
[20160804142412315]: /images/20220603/faffc4b1fa334d0f84e2566348ae50e3.png
[20160804142520861]: /images/20220603/a7c3a4826657402aa79a9fa9be441b23.png
[20160804142856300]: /images/20220603/191d245645284af18b955007cf207a27.png
[20160804142902995]: /images/20220603/8304031d5cf248e286db9f9e4d1e936a.png
[20160804143758054]: /images/20220603/4a0fcfa1c6eb4ae3b24e0349b60e71b3.png
[20160804151721930]: /images/20220603/10757206b42a44c4a644200b159a5306.png
[20160804151747515]: /images/20220603/da2413fd518a47eb9760fcdb36681992.png