Safety Helmet Detection

Objective: The objective of the project is to create a realtime object detection model for safety helmet.

Live Demo : https://helmet-detection-frontend-hcwvxl5l1-deepakdpyqaz.vercel.app/

Use Cases

• Can be used to detect safety helmet at construction sites
• Can be used at mining sites
• Model can be trained further to detect normal helmets as well

Tools Used

• Object Detection: Yolov4 and tiny-yolov4
• Front-end: Next js
• Back-end: Flask

Dataset Used: Safety Helmet Detection

Models Trained:

1. YoloV4 model

• Model Size: 244.2MB
• Speed (in FPS for realtime prediction): 1-2 FPS (On CPU)
• Accuracy: 94.91%

2. Tiny-yoloV4 model

• Model Size: 22.4MB
• Speed (in FPS for realtime prediction): 10-15 FPS (On CPU)
• Accuracy: 83.36%
• Demo

tiny_yolo_demo.mp4

Observation and Result

From the comparative observation, we can say that Yolov4 model is better in terms of accuracy but tiny-yolov4 is better in terms of model size and speed and only a little less in accuracy, so for deployment on web, tiny-yolo model is a better option as it will be light weight and work fast