Project-Template for IW276 Autonome Systeme Labor

After former projects already focused on identifying objects in images and video feeds this project is about re-identifying those objects in a sequence of images.

Demo Video

Presentation Video

This work was done by Sascha Isele, Michael Niepalla, Kai Michael Schwark during the IW276 Autonome Systeme Labor at the Karlsruhe University of Applied Sciences (Hochschule Karlruhe - Technik und Wirtschaft) in WS 2021 / 2022.

Table of Contents

• Requirements
• Prerequisites
• Docker
• Running
• Acknowledgments

Requirements

• Python 3.6 (or above)
• OpenCV 4.1 (or above)
• Jetson Nano
• Jetpack 4.4

[Optional] Monitor, mouse & keyboard directly connected to the Jetson Nano

Prerequisites

1. Clone the repository

// https
git clone https://github.com/IW276/IW276WS21-P20.git

// ssh
git clone git@github.com:IW276/IW276WS21-P20.git

2. Move inside the directory

cd IW276WS21-P20

3. Enable jetson_clocks to increase performance

sudo jetson_clocks

4. (optional) using own images and detectionfile

If you intend to use your own data you have to put the images into the "/dataset" folder of the repository. Best is to put the images into for a folder called "/dataset/img" for example and the detection file into "/dataset/det" to have it seperated.

If you already have such kind of folder structure just copy it using the following command and insert te correct paths.

cp -a /source/. /dest/

if you dont have the folder structure as mentioned above add the folders using

mkdir /new-directory-path/

[Optional] If you intend to view the process live you have to use a monitor directly connected to the Jetson Nano, else there will most likely be an error telling you "could not connect to display / Aborted (core dumped)"

Docker

Executing and building the project is done using docker.

Build

Execute the following code in order to build the docker image (while being in the project folder /IW276WS21-P20)

sudo bash docker-build.sh

Running

To run the demo, pass path to the pre-trained checkpoint and camera id (or path to video file):

sudo bash docker-run.sh

In the Docker container run the three commands listed below together with one of the 4 options listet below.

cd /deep-person-reid
python3 setup.py develop
cd /src

Option 1: run application using builtin sample data with gui

python3 opencv_gui.py --image_path --detection_path --identified_images

Option 2: run application using builtin sample data headless

mkdir /images_outputfolder
python3 opencv_gui.py --identified_images /images_outputfolder

Option 3: run application using own data with gui

python3 opencv_gui.py --image_path IMAGES_FOLDER_LOCATION_PATH --detection_path DET_FILE_PATH

INFO: own images and detection file need to be placed in the /datasets folder previous to building.

Option 4: run application using own data headless

mkdir /images_outputfolder
python3 opencv_gui.py --image_path IMAGES_FOLDER_LOCATION_PATH --detection_path DET_FILE_PATH --identified_images /images_outputfolder

INFO: own images and detection file need to be placed in the /datasets folder previous to building.

INFO: for option 1 and 3 remember to have a monitor connected to you Jetson Nano.

Acknowledgments

This repo is based on

• deep-person-reid
• L4T-ML

Thanks to the original authors for their work!

Contact

Please email mickael.cormier AT iosb.fraunhofer.de for further questions.