Below PPE objects can be detected from the required video/image
- hard hat
- vest
- mask
- boots
PPE object detection after Yolo training with custom dataset:
Custom Dataset was prepared to extract images with PPE objects from Google.
Code link - https://github.com/rameshveer/YoloV3_PPE_prediction/blob/master/image_scrapper.ipynb
Bounding Boxes(BBox) for the objects in extracted images were manually done using below tool.
This tool will automatically calculate the x,y co-ordinates of the BBox and put it in yolov3 format via text file.
After collecting and annotating around 3500 PPE images, they were cleaned and fed to the yolov3 model.
Model was trained using Google Colab(Tesla T4 GPU) for 50 epochs only in 5 hrs. More epochs would fetch far more accurate results.