Yolov5 for Fire Detection

This project presents the development and implementation of an advanced Smart Water Sprinkler System designed for fire detection and extinguishing. Using and integrating artificial intelligence (AI), robotics, and high-pressure sprinkler technology, the system aims to address critical limitations in current firefighting methodologies. The AI component employs a model trained on a diverse dataset to detect fires accurately, while a precision-focused robotic arm precisely points to the source after getting the location of the fire. The high-pressure sprinkler system ensures rapid and efficient extinguishing.

Introduction
System Overview
Our Approach
Risk Assessment
Testing and Implementation
- Components and Costs
Conclusion

Introduction

This project aims to create a Smart Water Sprinkler System capable of detecting and extinguishing fires using a combination of AI, robotics, and high-pressure water sprinklers. By leveraging advanced technologies, this system provides a more efficient and effective solution to fire detection and suppression, potentially reducing property damage and saving lives.

System Overview

The Smart Water Sprinkler System consists of four main components:

Fire Detection and Localization Module: Utilizes a Raspberry Pi and a camera module running YOLOv5 for real-time fire detection and localization.
Robotic Arm: A 2 DOF arm equipped with servo motors for precise targeting of the fire source.
NodeMCU32-S: Controls the motor movements and handles remote operations.
High-Pressure Water Sprinkler: Ensures rapid extinguishing of detected fires.

The system is designed to operate in both manual and automatic modes, with IoT integration for seamless switching.

Our Approach

System Design

The system continuously monitors the environment using a Raspberry Pi camera mounted on a servo motor. The camera sweeps the area in 5-degree increments to detect fires. A thermal imager validates the presence of fire. Upon detection, the system calculates the precise location of the fire using image coordinates and directs the robotic arm accordingly. For testing, the system is enclosed in an 80x80x40 cm space, simulating a room.

Software Design

The software component involves two main models:

YOLOv5: Trained to detect fires and provide bounding box coordinates. It is optimized for real-time performance on a Raspberry Pi.
Random Forest Model: Aggregates sensor data to confirm the presence of a real fire, reducing false positives.

Mechanical Design

The robotic arm has 2 DOF, with one joint for horizontal movement and another for vertical movement. It uses MG995 servo motors for precise positioning. The arm is equipped with a water hose connected to a high-pressure pump for fire suppression.

Electrical Design

The electrical system includes a Raspberry Pi connected to a PI Camera and NodeMCU32-S. The NodeMCU32-S controls the servo motors, thermal imager, and smoke sensor. The system uses a computer power supply for powering all components and a relay for controlling the water pump.

Risk Assessment

Identified risks and their mitigation strategies are outlined in the table below:

Risks	Description	Likelihood	Impact	Mitigation Strategy
R1: Fire detection system failure	Failure to detect fire	Moderate	High	Regular testing, backup sensors
R2: Robotic arm malfunction	Arm fails to operate correctly	Low	Moderate	Regular inspection, emergency manual control
R3: Water pump failure	Pump fails to operate	Low	High	Regular maintenance, backup pump
R4: Electrical system failure	Electrical components fail	Moderate	High	Surge protection, backup power supply
R5: False positive fire detection	Incorrect fire detection	Moderate	Moderate	Verification system, additional sensors
R6: Inadequate water supply	Insufficient water for extinguishing	Low	High	Regular monitoring, low water level alarms
R7: Integration issues	Software and hardware integration issues	Moderate	Moderate	Thorough testing, regular updates
R8: Inaccurate positioning	Robotic arm misalignment	Moderate	Moderate	Feedback mechanisms, regular calibration
R9: Safety risks	High-pressure water hazards	Low	High	Safety mechanisms, user training

Testing and Implementation

Components and Costs

A detailed list of components and their costs is provided below:

Component	Part Number	Link	Price
Smoke Detector	MQ-2 Smoke Gas Sensor	MQ-2 Smoke Gas Sensor	$3.00
Camera	Raspberry Pi B/B+ Official Camera	Raspberry Pi Camera	$5.50
Robotic Arm	2DOF Robot Arm with 3 MG995 Motors	Robot Arm	$55.00
Casing	Plexi Glass	N/A	$100.00
Raspberry Pi	Raspberry Pi 3 Model B+	Raspberry Pi 3 B+	$55.00
Water Pump	12V Fuel/Water Pump	Water Pump	$25.00
Arduino Relay	Relay 5V 10A	Relay	$0.60
Power Supply	Computer Power Supply 12V,5V,3.3V	Power Supply	$10.00
NodeMCU-32s	NodeMCU-32S	NodeMCU-32S	$7.00

For more information on parts and components, see Appendix C.

Conclusion

The Smart Water Sprinkler System represents a significant advancement in fire detection and suppression technology. By integrating AI, robotics, and IoT capabilities, this system offers a reliable and efficient solution for enhancing fire safety in various settings. Continued testing and development will ensure the system meets all safety and performance standards.

Credits to spacewalk01 for the fire detection model

Install

Clone this repo and use the following script to install YOLOv5.

# Clone
git clone https://github.com/spacewalk01/Yolov5-Fire-Detection
cd Yolov5-Fire-Detection

# Install yolov5
git clone https://github.com/ultralytics/yolov5
cd yolov5
pip install -r requirements.txt

Training

I set up train.ipynb script for training the model from scratch. To train the model, download Fire-Dataset and put it in datasets folder. This dataset contains samples from both Fire & Smoke and Fire & Guns datasets on Kaggle. I filtered out images and annotations that contain smokes & guns as well as images with low resolution, and then changed fire annotation's label in annotation files.

python train.py --img 640 --batch 16 --epochs 10 --data ../fire_config.yaml --weights yolov5s.pt --workers 0

Prediction

If you train your own model, use the following command for detection:

python detect.py --source ../input.mp4 --weights runs/train/exp/weights/best.pt --conf 0.2

Or you can use the pretrained model located in models folder for detection as follows:

python detect.py --source ../input.mp4 --weights ../models/best.pt --conf 0.2

Results

The following charts were produced after training YOLOv5s with input size 640x640 on the fire dataset for 10 epochs.

P Curve	PR Curve	R Curve

Prediction Results

The fire detection results were fairly good even though the model was trained only for a few epochs. However, I observed that the trained model tends to predict red emergency light on top of police car as fire. It might be due to the fact that the training dataset contains only a few hundreds of negative samples. We may fix such problem and further improve the performance of the model by adding images with non-labeled fire objects as negative samples. The authors who created YOLOv5 recommend using about 0-10% background images to help reduce false positives.

Ground Truth	Prediction

Feature Visualization

It is desirable for AI engineers to know what happens under the hood of object detection models. Visualizing features in deep learning models can help us a little bit understand how they make predictions. In YOLOv5, we can visualize features using --visualize argument as follows:

python detect.py --weights runs/train/exp/weights/best.pt --img 640 --conf 0.2 --source ../datasets/fire/val/images/0.jpg --visualize

Input	Feature Maps

Reference

I borrowed and modified YOLOv5-Custom-Training.ipynb script for training YOLOv5 model on the fire dataset. For more information on training YOLOv5, please refer to its homepage.

nizarsh98 / smartfireextinguisher Goto Github PK

smartfireextinguisher's Introduction