基于faster r-cnn的车辆跟踪检测和计数

[AI开箱]Implementation of Countor: count without bells and whistles

https://github.com/dyh/unbox_AndresOsp_Track1
https://www.bilibili.com/video/BV1Ja4y1s7mh

---

Unbox AI

• unbox opensource projects and products of Artificial Intelligence

welcome to subscribe my channel

• youtube channel
• bilibili channel

unbox project

• Implementation of Countor: count without bells and whistles

video

• youtube

intro

• vehicle counting, vehicle tracking
• top teams in the 2020 AI City Challenge
• Track 1: Multi-Class Multi-Movement Vehicle Counting

system requirements

• ubuntu 18.04
• python >= 3.6
• cuda 10.2

setup environments

1. clone source codes

   $ git clone https://github.com/dyh/unbox_AndresOsp_Track1.git

2. enter project directory

   $ cd unbox_AndresOsp_Track1

3. create a python virtual environment

   $ python3 -m venv venv

4. activate the virtual environment

   $ source venv/bin/activate

5. upgrade pip

   $ python -m pip install --upgrade pip

6. install requirements package

   1. install other packages

      $ pip install -r requirements.txt

   2. install NVIDIA DALI (based on cuda and pytorch)

      and you could choose other version at here

      $ pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali-cuda100

execute the program

1. download the weigths from the pytorch model zoo

   wget https://download.pytorch.org/models/fasterrcnn_resnet50_fpn_coco-258fb6c6.pth

2. detect the “./unbox_test/input/MVI_40855.mp4” video file of this project

   $ python unbox.py

3. the output results are saved in “./unbox_test/output” directory for some image files

   0.png, 1.png, 2.png … n.png

4. we can use ffmpeg to merge these images into one video file

   $ ffmpeg -f image2 -i ./unbox_test/output/%d.png ./unbox_test/output.mp4

sample dataset

• the test video MVI_40855.mp4 is made up of images in the directory “MVI_40855” in the DETRAC-test-data.zip file.
• download dataset: http://detrac-db.rit.albany.edu/download
• introduction of dataset: http://smart-city-sjsu.net/AICityChallenge/data.html

---

AI开箱

• 人工智能开源项目和产品开箱

欢迎订阅我的频道

• bilibili频道
• youtube频道

开箱项目

• Implementation of Countor: count without bells and whistles

视频

• bilibili

简介

• 车辆计数, 车辆追踪
• top teams in the 2020 AI City Challenge
• Track 1: Multi-Class Multi-Movement Vehicle Counting

系统需求

• ubuntu 18.04
• python >= 3.6
• cuda 10.2

环境配置

1. 下载代码

   $ git clone https://github.com/dyh/unbox_AndresOsp_Track1.git

2. 进入目录

   $ cd unbox_AndresOsp_Track1

3. 创建python虚拟环境

   $ python3 -m venv venv

4. 激活虚拟环境

   $ source venv/bin/activate

5. 升级pip

   $ python -m pip install --upgrade pip

6. 安装软件包

   1. 安装其他包

      $ pip install -r requirements.txt

   2. 安装 NVIDIA DALI (基于 cuda 和 pytorch)

      你也可以在 这里 选择其他版本

      $ pip install --extra-index-url https://developer.download.nvidia.com/compute/redist nvidia-dali-cuda100

运行程序

1. 从pytorch model zoo下载weigths文件

   wget https://download.pytorch.org/models/fasterrcnn_resnet50_fpn_coco-258fb6c6.pth

2. 对项目中的 ./unbox_test/input/MVI_40855.mp4 视频文件进行检测

   $ python unbox.py

3. 输出结果为图片文件，保存在 ./unbox_test/output 目录

   0.png, 1.png, 2.png … n.png

4. 可以使用 ffmpeg 将图片文件合并为视频文件

   $ ffmpeg -f image2 -i ./unbox_test/output/%d.png ./unbox_test/output.mp4

样本数据集

• 测试视频 MVI_40855.mp4 是由 DETRAC-test-data.zip 文件中的 MVI_40855 目录下图片组成。
• 数据集下载 http://detrac-db.rit.albany.edu/download
• 数据集说明 http://smart-city-sjsu.net/AICityChallenge/data.html

---

below is the origin README file

---

forked from AndresOsp/Track1

Implementation of Countor: count without bells and whistles

This repository contains our implementation for 2020 AICity Challenge, and we achieve second place in Track 1: Multi-Class Multi-Movement Vehicle Counting.

Abstract:

The effectiveness of an Intelligent transportation system (ITS) relies on the understanding of the vehicles behaviour. Different approaches are proposed to extract the attributes of the vehicles as Re-Identification (ReID) or multi-target single camera tracking (MTSC). The analysis of those attributes leads to the behavioural tasks as multi-target multi-camera tracking (MTMC) and Turn-counts (Count vehicles that go through a predefined path). In this work, we propose a novel approach to Turn-counts which uses a MTSC and a proposed path classifier. The proposed method is evaluated on CVPR AI City Challenge 2020. Our algorithm achieves the second place in Turn-counts with a score of 0.9346.

Get started:

• Clone this repository

• Download the weigths from the pytorch model zoo:
  wget https://download.pytorch.org/models/fasterrcnn_resnet50_fpn_coco-258fb6c6.pth

• Download and extract the dataset from AI City Challenge

  • The dataset must contain:
    • A Folder with all the videos
    • The list_video_id.txt file witch gives the ID of each video
    • The track1_vid_stats.txt file with the number of frames of each video

• Open the config.py file:

  • Set the paths for the corresponding files
  • Set the GPU numbers that will be used to run the algorithm
  • If you run out of memory or problems with the processing power of the GPU please set parallel_processes to 1

• Then just run ‘python run.py’

• When the program finish the results are in the Results folder.

Dependencies

• pytorch
• torchvision
• nvidia-dali
• numpy
• pandas
• Pillow
• pyyaml

Citation

@InProceedings{Ospina_2020_CVPR_Workshops,
author = {Ospina, Andres and Torres, Felipe},
title = {Countor: Count Without Bells and Whistles},
booktitle = {The IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2020}
}

Authors:

• Andres OSPINA
• Felipe TORRES

Used repositories

We used code from the following repositories:

• Tracking without bells and whistles
• Longest Processing Time

License

See LICENSE. Please read before use.