Lume

• 

  Original Model

• 

  Fine-Tuned Model

• 

  Model Graphs before Fine-Tuning

• 

  Model Graphs After Fine-Tuning

• 

  Comparison Between Original and Fine-Tuned YOLO V8 Model

• 

  Smoke Detection

Inspiration

The LA Wildfires have wreaked havoc on the local population. Throughout the world too, climate change has meant an unprecedented quantity of wildfires. These fires have cost thousands of lives, damage to properties and wildlife and are responsible for one-third of global CO2 emissions! Thus, for prevention and better crisis management, multi-point detection of smoke and wildfire is crucial

What it does

This model detects smoke and fire in images and videos using AI and alerts the relevant authorities to help prevent and manage crisis.

How we built it

We built this wildfire detection system using YOLO for fire detection and OpenCV for video processing. The model analyzes video frames in real time and highlights detected fires with bounding boxes. We use an API key to call 911 in case a fire is detected. If we press No, and it continues to detect a fire, then it will proceed to keep prompting you. We used Flask, SCRIPTING in JS and HTML and CSS to work on the dashboard.

Challenges we ran into

The YOLOv8 model had false detections which had to be resolved by fine tuning the model, this involved preprocessing the dataset through data augmentation techniques like color space transformation. We had issues with installing the dependencies for flask, due to incompatibility with the python version.

Accomplishments that we're proud of

Using API key to call 911 incase of fire detection. Accurately carrying out multi point detection, giving higher confidence scores in regions of high fire intensity.

What we learned

Integrating api keys to make calls to emergency services after smoke is detected for 5 straight seconds in 5 different video frames. We learned to collaborate with a team as we had previously only worked on projects separately. We were able to fine-tune a model by using different data augmentation techniques and we were able to

What's next for Lume

We want to be able to implement this complementing CCTV cameras to be able to monitor wildfires. In doing so, smoke/fire detection would also alert the relevant authorities and help in crisis management. We want to use Google Maps API to access and thus share the exact location of the fire upon detection. Make a working website where you can upload media and get an output showing the possibility of a wildfire.

Built With

• css
• html
• javascript
• jupyter
• opencv
• python
• roboflow