Pedestrian detection Resources

(reproduced, did not find the original source, such as invasion of the deletion or reminder to increase the source, thank you.) )

Pedestrian detection resources (top) Review literature http://www.cvrobot.net/pedestrian-detection-resource-1-summary-review-survey/

Pedestrian detection resources (below) code data http://www.cvrobot.net/pedestrian-detection-resource-2-code-and-dataset/

Pedestrian detection has an extremely wide range of applications: Intelligent assisted driving, intelligent monitoring, pedestrian analysis, and intelligent robotics. Since 2005, pedestrian detection has entered a rapid development stage, but there are still many problems to be solved, mainly in terms of performance and speed can not reach a trade-off. In recent years, the development of automated driving technology, led by Google, is in full swing, and there is an urgent need to be able to quickly and effectively detect pedestrians, in order to ensure the safety of pedestrians during automatic driving will not pose a threat. 1 status of pedestrian detection

Probably can be divided into two categories 1.1 based on background modeling

By using the background modeling method, the target of foreground motion is extracted, the feature is extracted in the target area, and then the classifier is used to classify and determine whether the pedestrian is included, and the main problems existing in background modeling are as follows:

1) must adapt to changes in the environment (such as changes in the illumination of the image color changes);

2) Jitter of the camera (such as the movement of a handheld camera when taking pictures);

3) Dense objects in the image (e.g., dense objects such as leaves or trunks) should be detected correctly;

4) must be able to correctly detect the background object changes (such as the new stop car must be in a timely manner to the background object, and the stationary start moving objects need to be detected in a timely manner).

5) object detection often appear in the Ghost area, Ghost area refers to when a stationary object began to move, the back of the static error detection algorithm may be the original object covered by the detection of the area of the fault as the motion, this area becomes ghost, Of course, the original moving object into a static will also introduce the Ghost region, Ghost region in the detection must be eliminated as soon as possible. 1.2 method based on statistical learning

This is also the most commonly used method of pedestrian detection, based on a large number of samples to build pedestrian detection classifier. The extracted features mainly include gray, edge, texture, color, gradient histogram and other information of the target. Classifiers mainly include neural networks, SVM, AdaBoost, and deep learning, which is now considered beloved by computer vision.

The present difficulties in statistical learning:

1) Pedestrian posture, costumes vary, complex backgrounds, different pedestrian scales and different lighting environments.

2) The distribution of the extracted features in the feature space is not compact enough;

3) The performance of the classifier is greatly influenced by the training samples;

4) Negative samples from offline training cannot cover all real-world scenarios;

The current pedestrian detection is basically based on the HOG+SVM Pedestrian detection algorithm published by French researchers Dalal in the 2005 CVPR (histograms of oriented gradients for Human Detection, Navneet Dalel,bill Triggs, CVPR2005). HOG+SVM as a classic algorithm is also integrated into the OPENCV inside, can be directly called to implement pedestrian detection

In order to solve the speed problem, we can use the background difference method to study pedestrian detection, the premise is that the method of background modeling is effective enough (i.e., the effect is good speed), and the method of obtaining better detection results is usually based on multi-feature fusion and cascade classifier. (common features include harry-like, hog features, LBP features, Edgelet features, CSS features, cov features, integration channel features, and centrist features). 2 Articles of the review class 2.1 Pedestrian Detection 10 year review

Ten years of pedestrian Detection, what has We learned?

A 2014 ECCV article is a review of the development of pedestrian detectiond over the past ten years, analyzing the methods proposed by over 40 papers in the last 10 years from the perspective of Dataset,main approaches, The effect of improving the complexity of feature is also evaluated.

Download: http://rodrigob.github.io/documents/2014_eccvw_ten_years_of_pedestrian_detection_with_supplementary_material.pdf

Study Note: http://blog.csdn.net/mduke/article/details/46582443
a review on 2.2 p.dollar Pami 2012

P.dollar, C. wojek,b Schiele, et al pedestrian detection:an evaluation of the state of the art [J]. IEEE Transactions on Pattern analysis and Machine Intelligence, 2012, 34 (4): 743-761.

A summary article on pedestrian detection, published in Pami in 2012, a total of 20 pages, a simple description of the common 16 pedestrian detection algorithms, and tests on 6 public test libraries, gives the advantages and disadvantages of various methods and their applicability. In addition, the development direction and trend of pedestrian detection in the future are pointed out.

Download: http://vision.ucsd.edu/~pdollar/files/papers/DollarPAMI12peds.pdf 2.3 CVPR2010 Hof and CSS

Https://www.d2.mpi-inf.mpg.de/CVPR10Pedestrians

New Features and Insights for pedestrian Detection

This article uses the improved hog, the HOF and CSS (color self similarity) features, using the Hik SVM classifier. The author of this paper is the Germans: Stefen Walk. Stefan Walk is currently teaching at the Federal Polytechnic University in Zurich. 2.4 Integral Channel Features

California Institute of Technology 2009-year pedestrian detection article: Integral channel Features (integral channels feature)

This article is the same author as the 2012 Pami review article. Author: Piotr Dollar

Paper Download: http://pages.ucsd.edu/~ztu/publication/dollarBMVC09ChnFtrs_0.pdf

Chinese Note: http://blog.csdn.net/carson2005/article/details/8455837 2.5 The fastest pedestrian Detector in the West

Dollar in 2010 Bmvc's "The fastest pedestrian detector in the West" proposed a new idea, this idea only needs to train a standard model, detection n/k (K≈10) and then the rest of the The characteristics of n-n/k-sized images do not require this complex calculation, but are based on the results of this n/k, and are estimated by another simple algorithm, the basis for this idea is that the characteristics of similar images can be accurately estimated enough.

Download: http://vision.ucsd.edu/sites/default/files/FPDW_0.pdf 2.6 DPM algorithm for target detection

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 the author research the DPM algorithm to do target detection of the article, the source code can be downloaded.

Author's profile: http://cs.brown.edu/~pff/papers/ 2.7 Using the DPM model to detect adhesion

Detection and Tracking of occluded people

IJCV2014 article, the use of DPM model, the detection of adhesion is very serious pedestrian, the effect is very good.

Download: http://www.bmva.org/bmvc/2012/BMVC/paper009/
2.8 UDN Algorithm

ICCV2013: