NOTE: This is a basic prototype where only the smoky and non smoky vehicles are classified, need to add a ton of functionality
This project aims to detect vehicles emitting black or gray exhaust gases, which indicate a faulty exhaust system. It uses transfer learning applied to a ResNet-152 model pre-trained on the ImageNet dataset to differentiate between smoky and non-smoky vehicles.
Personal note: "i was very very interested in automating things that improve civic sense and anything that the govn should do (because they don't do shit). trying to capture vehicles that produce the smoke which is bad for the environment but also makes the vehicles behind a pain to be behind and take risky overtakes to get ahead of such smoky vehicles. this is literally my first ML project!"
Faulty vehicle exhausts are a significant contributor to air pollution and can cause various health problems. By detecting such vehicles, this project aids in environmental protection efforts and helps to enforce vehicle emission standards.
The model was trained on a custom dataset of 400 images labeled as 'smoky' or 'non-smoky'. This was created synthetically by https://github.com/srimantacse/VehicleSmokeDataset To implement this project in future, need to make dataset from real world.
The model is based on ResNet-152, a deep residual network with 152 layers. We have done Transfer-Learing on the Resnet model with the smoky vehicle dataset to adapt the model to this specific task.
Clone the repository to your local machine:
git clone https://github.com/your-username/smoky-vehicle-detection.git cd smoky-vehicle-detection
Python , PyTorch, PIL, etc.
python train_model.py
Have secured 91% accuracy on the test set
Have to add multiple more features such as number plate detection, mapping the number plate with the vehicle detected. This is going to get quite complex to accomplish, will try to add these funcitonalities in the future
open to discuss in detail how we can make it production ready by how we could implement the future work