Pedestrian detection 3 (Overview of the latest paper on Pedestrian detection)

Serial number	Introduction	Thesis	Source
0	Summary of Pedestrian detection in PAMI in 2012: Pedestrian detection an evaluation of the state of the art Piotr dollar This article compares many of the latest Pedestrian detection algorithms.. This paper is referred to as pami2012	Pedestrian detection an evaluation of the state of the art
1	Pami2012: New Features and insights for pedestrian detection The improved hog, namely, Hof and CSS (color self similarity) features are used in this article, and hik SVM classifier is used. The author of this article is German: Stefen walk. Stefan walk currently teaches at Zurich federal Polytechnic University.	New Features and insights for pedestrian detection	Https://www.d2.mpi-inf.mpg.de/CVPR10Pedestrians
2	In pami2012's summary, the top 2nd algorithms are: Department of California Institute of Technology's Pedestrian detection in 2009: integral channel features) This article is the same as the PAMI summary article in 2012. Piotr dollar	Integral channel features	Http://www.vision.caltech.edu/Image_Datasets/CaltechPedestrians/ Library and Demo code for various Pedestrian detection The Matlab code contains the complete source code for training and testing algorithms. The code in the compressed package contains the algorithms mentioned in almost all the papers of the author. The code of the author's latest pami2014 thesis is also included in the compressed package.
3	Pami2012: Ranking 3rd Algorithms The fastest pedestrian detector in the West This article is the same as the PAMI summary article in 2012. Piotr dollar	The fastest pedestrian detector in the West	Author's homepage: Http://vision.ucsd.edu /~ Pdollar/research.html The Matlab code download page for calculation in this article: Http://vision.ucsd.edu /~ Pdollar/toolbox/doc/index.html
4	AuthorPiotr dollarArticles on Pedestrian detection written in 2009	Pedestrian detection a benchmark.
5	Cvpr2008: A discriminatively trained, multiscale, deformable part model   Pami2010: Object Detection with discriminatively trained Part Based Models   Cvpr2010: Cascade object detection with deformable part Models   The above three articles are all articles by the author studying the DPM Algorithm for target detection. You can download the source code. This algorithm is not mentioned in the pami2012 summary. I don't know why.	A discriminatively trained, multiscale, deformable part model   Object Detection with discriminatively trained Part Based Models   Cascade object detection with deformable part Models	Author's personal homepage: Http://cs.brown.edu /~ Pff/papers/
6	In the 2014 article of ijcv, the DPM model was used to detect pedestrians with severe adhesion conditions and the results were very good.	Detection and Tracking of occluded people	Currently, the source code related to this paper cannot be found.
7	Iccv2013: In short, it is called The UDN algorithm. From the detection effect described in this article, this method is the best in all methods, and the effect far exceeds that of other methods. After studying the thesis and the algorithm source code, this algorithm is used with the author in another paper. In addition, the algorithm is used for image scanning to obtain the rectangle frame, then we use this method to further confirm the rectangle frame and reduce the false alarm rate and false alarm rate. Other papers: Multi-Stage contextual deep learning for pedestrian detection   This article does not contribute much to pedestrian detection. Just use CNN of deep learning to confirm the candidate window. The main pedestrian detection algorithm is hog + CSS + AdaBoost.	Joint deep learning for pedestrian detection   Multi-Stage contextual deep learning for pedestrian detection	The Chinese University of Hong Kong, joint deep learning for pedestrian detection, resources related to pedestrian detection papers: Http://www.ee.cuhk.edu.hk /~ Wlouyang/projects/ouyangwiccv13joint/index.html
8	Eccv papers in 2010: Multiresolution models for Object Detection The algorithm described in this article is quite effective, but the author has not published the source code. I don't know whether the results in this paper are true or not.	Multiresolution models for Object Detection	The multires algorithm detects pedestrians. The author's personal homepage: Http://www.ics.uci.edu /~ Iypark/ The author has not published the source code or the demo.
9	Iccv papers in 2009, the detection results andPiotr dollarThe effect is comparable. The author only published the test software and did not published the source code. This feature is similar to the centrist feature and can describe the global contour of the human body.	An HOG-LBP human detector with partial occlusion handling	Http://www.xiaoyumu.com/ Http://vision.ece.missouri.edu /~ Wxy/index.html Http://web.missouri.edu /~ Hantx/
10	Using the centrist feature, the centrist feature is improved by the feature of the guid. The authors compared the centrist features with those of hog and HSV features to demonstrate that centtrist features have a good effect in describing pedestrians. The author is a doctor from Nanyang, a Chinese researcher. In my personal understanding, there is not much innovation in the characteristics of centrist, and there is no big difference between it and traditional Chinese medicine. The author himself also said in the article that the algorithm is not as effective as hog and HSV, but it is only faster.	Real-time human Detection Using Contour cues	The source code only contains the test source code and does not contain the code for training classifier. Http://www.c2i.ntu.edu.sg/jianxin/projects/C4/C4.htm

 

In general, there are three best methods in these latest articles:

1) Improved hog + improved SVM. That is, the method in pami2012's No. 1 paper. Unfortunately, the source code cannot be found.

2) Hof + CSS + AdaBoost .. That is, the second method in pami2012. Find the MATLAB source code.

3) hog + HSV + SVM method. That is, the method in the paper with the serial number 9 in the above table. No source code.

4) DPM. That is, the method in serial number 5 and 6 in the above table has the source code.

Original article: http://blog.csdn.net/dpstill/article/details/22420065

Pedestrian detection 3 (Overview of the latest paper on Pedestrian detection)