SafetyNet

Industrial Safety is essential for promoting worker safety, preventing accidents, ensuring compliance, and enhancing overall operational efficiency in industrial environments. This project aims to develop a real-time video analytics tool that enhances industrial safety by detecting and preventing potential hazards and unsafe situations in industrial environments.

Video Demo

Objective

The objective of this project is to leverage Deep Learning techniques to monitor live video feeds and provide real-time hazard detection, prompt alerts, and interventions, enhancing industrial safety and preventing accidents in industrial environments.

Project Structure & Flow

Technologies & Tools

• Python
• OpenCV
• Tensorflow
• YOLO (You Only Look Once) v8
• Flask (APIs and Back-end)
• React (Web-Application / Front-end
• Firebase - Firestore (NoSQL Database)
• RaspberryPi (Sensor Integration)

Features

• Real-time video analysis for hazard detection.
• Proactive alert system for immediate intervention.
• User-friendly web interface for monitoring and control.
• Integration with existing industrial video data.
• Dashboard for factory managers to track safety statistics and incidents.

Getting Started

Follow these steps to get started with the project:

1. Clone this repository to your local machine:

   git clone (https://github.com/smartinternz02/SBSPS-Challenge-10024-SafeZone-Real-time-Video-Analytics-for-Industrial-Safety)https://github.com/smartinternz02/SBSPS-Challenge-10024-SafeZone-Real-time-Video-Analytics-for-Industrial-Safety)
   cd SBSPS-Challenge-10024-SafeZone-Real-time-Video-Analytics-for-Industrial-Safety

2. Install the required dependencies:

   pip install -r requirements.txt

3. Start the application:

   cd flask_api
   python app.py

4. Run web application:

   cd frontend
   npm i -S react-scripts
   npm start

5. Default - Test Credentials for the application

   [email protected] / 123456
   [email protected] / 123456
   

Model statistics

1. Class labels:

[Hardhat, Mask, NO-Hardhat, NO-Mask, NO-Safety Vest, Person, Safety Cone,
Safety Vest, machinery, vehicle]

The model that was used in order to train this model was the yolov8l model that consists of 268 layers and 43,668,288 parameters. The model needs approximately 165 GFLOPS of compute power to run. The model was trained on a custom dataset with the aforementioned classes and the hardware that was required to train that model was: 2x Nvidia Tesla T4 GPUs with 16gb VRAM each, 30GB of RAM and 4v CPU compute Engine.

The model was trained for 310 epochs and took a total amount of time of nearly 36 hrs to train completely.

2. Model Metrics:

a. mAP50 (B):

The mAP50 B is the Mean Absolute Precision where it measures the Average Precision of detections that have at least a 50% overlap with ground truth objects while excluding those overlapping with the background.

b. mAP_0.5:

The mAP_0.5 is the same as the previous one but in contrast, mAP_0.5 calculates Average Precision based on detections with an Intersection over Union (IoU) of 0.5 or higher with ground truth objects.

c. Precision:

Precision in object detection is a metric that assesses the accuracy of a model's detections. It measures the ratio of correctly predicted positive detections to the total number of positive predictions made by the model. In the context of object detection, a "positive detection" refers to a bounding box or region proposed by the model that correctly corresponds to an actual object in the scene.

d. Recall:

Recall, in the context of object detection, is another important metric that evaluates the completeness of a model's detections. It measures the ratio of correctly predicted positive detections to the total number of actual positive objects present in the scene.

Usage

• Upload industrial video data for analysis.
• The system will analyze the video in real-time, detecting potential hazards.
• The dashboard provides real-time statistics on safety incidents.
• When a hazard is detected, the system triggers alerts and interventions as needed.

Outputs

1. Safety Gear Recognition:

2. Emergency Hand Gesture Detection:

Implementation Screenshots

Introduction

Login for Admins & Managers

Admin Dashboard

It consists of Organization wide data, i.e. data collected from all the factories / industrial sites are crunched together for the admin of the organization to observe.

The admin dashboard also has access to analytics and stats related to the safety of the industrial sites, the graphs of safety-scores, fire-incidents and harmful gas sensor alerts

Manager Dashboard

It consists of Factory / Industrial area wide data, i.e. data collected from that particular factory or industrial site is crunched together for the manager of the respective site to observe.

The manager dashboard also has access to analytics and stats related to the safety of the industrial sites, the graphs of safety-scores, fire-incidents and harmful gas sensor alerts