I can detect Pothole

Inspiration

As we know that for safe and better driving, a driver must avoid potholes. SO, I made a project which detects the potholes on roads. This will help the driver to avoid them while driving.

What it does

"I can detect Pothole" as the same suggests is a pothole prediction project which help in prediction of potholes to provide the drive safe drive experience.

This Project is based upon collection of images from phone application's and using Deep Learning Technique(Mask RCNN), we have trained a custom machine learning model for detection of potholes.

How we built it

• As a basic model for transfer learning, we employ EfficientDet-Lite[0-4], which is a collection of mobile/IoT-friendly object recognition models developed from the EfficientDet architecture. The dataset is divided into training and testing sets, with 90% of the data being used for training and 10% being utilized for validation and testing. For the model, the training phase was implemented with (300 epochs, batch size 32, and picture size 320320 pixels).

• For our labelled dataset and bespoke model, we used a modified YOLOv3 architecture with Darknet-53. We employ a Convolutional Neural Network (CNN) in the YOLO architecture to detect road flaws, since the positive highlighted regions reflect road problems. The dataset is divided into training and testing sets, with 85 percent of the data used for training and 15% for testing. On Google Col laboratory, we've trained many models with the aforementioned architecture and hyperparameters. The model was trained with (8000 epochs, batch size 64, and picture size 418 418 pixels) during the training phase.

Challenges we ran into

There were various challenges which I faced while making this project.

• reading the data from the dataset
• training model to predict the potholes in the road
• the trying took so much time.

Accomplishments that we're proud of

I am proud on my self that I was able to complete this project and learned about machine learning and deep learning, which will help me in the future.

What we learned

I learned about the following things.

• Machine learning and deep learning.
• CNN (Convolutional Neural Network)
• yolov5 architecture

What's next for I can detect Pothole

In future, I want to add a feature in which the user can plot the pothole in maps so that other drivers can see is and avoid them while driving.

Built With

• cnn
• deep-learning
• jupyter-notebook
• matplotlib
• numpy
• opencv
• python
• tensorflow
• yolov5