liangzimei / pedestrain_detection

Compare the pedestrain detection using Faster RCNN and RPN+BF

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Comparing the differences between Faster RCNN and RPN+BF in pedestrain detection

By 庄月清,常亦谦

Introduction

行人检测具有极其广泛的应用:智能辅助驾驶,智能监控,行人分析以及智能机器人等领域。随着深度学习的性能的优越性,将深度学习的方法应用到行人中以提高检测准确率。本工程分别采用Faster R-CNN和RPN+BF网络,对Caltech数据集进行训练和测试,并比较两者的结果。

This code has been tested on Ubuntu 16.04 with MATLAB 2014b and CUDA 7.5.

Citing

RPN+BF

@article{zhang2016faster,
  title={Is Faster R-CNN Doing Well for Pedestrian Detection?},
  author={Zhang, Liliang and Lin, Liang and Liang, Xiaodan and He, Kaiming},
  journal={arXiv preprint arXiv:1607.07032},
  year={2016}
}

Faster R-CNN

@article{ren15fasterrcnn,
Author = {Shaoqing Ren, Kaiming He, Ross Girshick, Jian Sun},
Title = {{Faster R-CNN}: Towards Real-Time Object Detection with Region Proposal Networks},
Journal = {arXiv preprint arXiv:1506.01497},
Year = {2015}
}

Requirements

  1. Caffe build for RPN+BF (see here)

    • If the mex in 'external/caffe/matlab/caffe_faster_rcnn' could not run under your system, please follow the instructions on our Caffe branch to compile and replace the mex.
  2. MATLAB

  3. GPU: Titan X, K40c, etc.

How to build and run

  1. Download the special caffe vision for this project(see here), and follow the readme.md in it to build and run.

  2. Download the annotations and videos in Caltech Pedestrian Dataset and put them in the three folder (videos|res|annotations) under ./RPN_BF/external/code3.2.1/data-USA and ./faster_rcnn_caltech/external/code3.2.1/data-USA.

  3. The ./faster_rcnn_caltech include the code of faster rcnn on caltech datasets, follow the readme.md to make sure it perform well. Start MATLAB from the repo folder, and Run script_faster_rcnn_caltech.m to train and test the faster rcnn on Caltech, script_fast_rcnn_caltech_eval.m to evaluate the result after train and test.

  4. The ./RPN_BF include the code of RPN+BF on caltech datasets, follow the readme.md to make sure it perform well, Start MATLAB from the repo folder, and Run script_rpn_pedestrian_VGG16_caltech to train and test the RPN model on Caltech, Run script_rpn_bf_pedestrian_VGG16_caltech to train and test the BF model on Caltech (the evaluation result is included in the test).

  5. Hopefully it would give the evaluation results.

Experiment results

Faster RCNN

image image image

In addition, we have raised the mr to 30% for Faster RCNN on the caltech datasets.

RPN+BF

image image image image

ezoic increase your site revenue

About

Compare the pedestrain detection using Faster RCNN and RPN+BF


Languages

Language:MATLAB 94.3%Language:C++ 3.0%Language:Cuda 2.2%Language:M 0.2%Language:Objective-C 0.2%