Pothole Detection in Roads using Pretrained Model and Cloud Computing
Suriya Narayanan R
Student, department of Artificial Intelligence and Data Science Bannari Amman Institute of Technology
Erode, India suriyarajavel02@gmail.com
Haripriya R
Assistant Professor, department of Computer Technology Bannari Amman Institute of Technolgy
Erode, India haripriyar@bistathy.ac.in
Abstract—The fundamental objective of this project is to make use of the advanced capabilities offered by technology that is cur- rently available on the market in order to locate and accurately identify potholes. The research identifies potholes in real-time video feeds obtained from vehicles by employing YOLOv8, a dependable pre-trained model for object detection. Utilising an Internet of Things (IoT) gateway, which creates a connection with Kinesis Video Streams in order to collect the video data, makes the procedure simpler and more efficient.The combination of the YOLOv8 algorithm with the Amazon SageMaker platform enables the accurate detection and localization of potholes within the frames of a particular film. The project makes use of the computing power made available by Amazon Web Services (AWS) in order to make the effective processing and analysis of video data inside a cloud-based environment possible.After the algorithm has determined that potholes are present, it will proceed to take precise notes of the coordinates that correlate to each individual hole. The data described above is gathered and saved within the environment provided by AWS, which provides helpful insights for operations relating to maintenance and repair. Because of the combination of pretrained models, cloud-based infrastructure, and Internet of Things (IoT) components, a dependable and scalable solution for the prompt identification of potholes has been produced. This has resulted in an improvement in both the safety of road networks and their maintenance.
Index Terms—pothole detection, deep learning, cloud comput- ing, object detection, pretrained models, convolutional neural networks
