helmet-project

Auto object detection at construction site using computer vision algorithm YOLO

Project for bachelor thesis

• Paper : pdf link
• PPT : ppt link
• Video : detecting video link

The project source code files are organized as below:

• construction-image-detecion-final.ipynb: Loads data from Google Drive to Colab environment, Preprocesses data and Trains model with yolov5 package. You should run this ipython notebook at Colab and your drive should contain folder Helmet-Project which contains ACID_6000.zip and Human-Body.zip
• detect.py: YOLOv5 package's object detecting module which is an open source. If you want to try object detection form Youtube video, type python detect.py --weights runs/train/yolo5s_construction_human/weights/best.pt --source 'https://youtu.be/MHaZXSneOGQ'

The project source data files are organized as below:

• ACID_6000: Contains 6,000 images of construction machines with annotation directories. The data is received from AIRCon-Lab, so you should contact Xiao, B ([email protected]) to get data
• Human-Body: Contains human gesture image with annotations. Label boxes' location and size is contained in MPHB-label-txt/MPHB-label.txt. You can download data from Multiple Pose Human Body Database

Run the ipynb file at Colab with data ready at Google Drive. Then you download the trained weights to yolov5/ directory and run detect.py module.

Youtube links used for detection:

1. Biggest Rc Construction-Site in the World! Rc Truck Action at Minibaustelle Alsfeld 2017!
2. Amazing Dangerous Fastest Excavator Operator Skill, Heavy Equipment Machines Truck Working Fail