Object_Detection_by_Infrared_ros

Robust Object Classification of Occluded Objects in Forward Looking Infrared (FLIR) Cameras using Ultralytics YOLOv3 and Dark Chocolate.And you can run this project in ROS.

Instructions

• Must have NVIDIA GPUs with Turing Architecture, Ubuntu and CUDA X installed if you want to reproduce results.

• Add the data provided by FLIR to a folder path called /coco/FLIR_Dataset.

• Place the custom pre-trained weights you downloaded from above into: /weights/*.pt

• Converted labels from Dark Chocolate are located in data/labels, which you unzipped above.

• The custom *.cfg with modified hyperparams is located in /scripts/cfg/yolov3-spp-r.cfg.

• Class names and custom data is in /scripts/data/custom.names and custom.data.

Downloads needed to run codebase

1. Download pre-trained weights here: link

2. FLIR Thermal Images Dataset: Download

3. Go into scripts/data folder and unzip scripts/labels.zip

4. Addt'l instructions on how to run Ultralytics Yolov3

Requirements

Python 3.5 or later with all requirements.txt dependencies installed

cd scripts
pip install -r requirements.txt

Install & Run Code:

• build messages

catkin build

• build cv_bridge Because ros cv_bridge don't compatible with Python3, you need to build cv_bridge with python3 in your workspace.

  you can reference this answer https://stackoverflow.com/questions/49221565/unable-to-use-cv-bridge-with-ros-kinetic-and-python3

• Run Code

Before you run this project, you need edit object_detection_by_camera.zsh line 8,

source ~/software/catkin_workspace/install/setup.zsh --extend

to your cv_bridge install path.

and run code

sudo chmod +x object_detection_by_camera.zsh
./object_detection_by_camera.zsh

Result