International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 10 Issue: 05 | May 2023
p-ISSN: 2395-0072
www.irjet.net
Pothole Detection Using ML and DL Algorithms 1
2
3
Mahesh Madhvi , Ruturaj Kondilkar , Dhiraj Bhakare , Bhargavi Kaslikar
4
Student, Final year B.Tech, Dept. of Electronics Engineering, K.J. Somaiya College Of Engineering, Mumbai, Maharashtra, India 4Professor, Electronics Department, K.J. Somaiya College of Engineering, Vidyavihar, Mumbai, Maharashtra, India ---------------------------------------------------------------------***-------------------------------------------------------------------1,2,3
Abstract - Potholes are a persistent problem in transportation infrastructure, causing vehicle damage and hazards for drivers. Manual detection methods are timeconsuming and costly. Computer vision and machine learning offer potential solutions for automating pothole detection. These technologies can improve road maintenance and safety by accurately identifying potholes. This paper presents an integrated approach for pothole detection and classification using YOLOv7, SVM, and segmentation techniques. A user- friendly web application is developed using Streamlit for easy interaction. The YOLOv7 model accurately detects and classifies potholes based on annotated training datasets. Segmentation techniques refine pothole regions, enabling precise boundary extraction for accurate analysis. The SVM model is trained on labeled pothole images to classify different pothole types. The web application allows users to upload images, detect and classify potholes, and visualize results. It aids road authorities and maintenance crews in identifying problematic areas and planning repairs, contributing to safer road conditions. Keywords: Computer vision, machine learning, pothole detection YOLOv7, SVM, segmentation techniques, web application, Streamlit, pre-processing, object detection, classification
1. INTRODUCTION Efficiently detecting and classifying potholes on roadways is vital for ensuring road safety and effective maintenance. This paper presents an integrated approach that utilizes the YOLOv7 object detection model for accurate pothole detection and classification. Additionally, a user-friendly web application is developed using Streamlit to provide an intuitive interface for interaction and result in visualization. The detection process relies on the YOLOv7 model, which is trained on annotated datasets to precisely locate and classify potholes in road imagery. The model's robust object detection capabilities make it well-suited for identifying potholes with high accuracy. To enhance user experience and facilitate result visualization, a web application is built using Streamlit. The application allows users to easily upload road images and applies the YOLOv7 model to detect and classify potholes. The detected potholes are then presented to the user in an
© 2023, IRJET
|
Impact Factor value: 8.226
|
accessible and user-friendly format. By combining the YOLOv7 object detection model and the Streamlit-based web application, our integrated approach provides an efficient solution for accurate pothole detection and classification. This empowers road authorities and maintenance crews to swiftly identify and address potholes, ultimately contributing to safer road conditions and more effective maintenance practices.
2. LITERATURE SURVEY In a study conducted [1], various image pre-processing and segmentation methods were explored to improve the accuracy of pothole detection. The study specifically implemented the difference between Gaussian-Filtering and clustering-based segmentation methods and compared their results. The findings revealed that the K- Means clustering-based segmentation method was the most efficient in terms of computing time, while edge detection-based segmentation was the most accurate. The study aimed to identify a superior method for pothole detection compared to traditional methods, and this objective was achieved by using performance measures to evaluate the different techniques reviewed. Their research paper [2] discussed a pothole detection model using computer vision and machine learning. The study involved collecting road images from BMC Mumbai and applying various computer visionoperations such as pre-processing, morphological operations, canny edge detection, and decision tree algorithms to detect potholes. The proposed model aimed to identify potholes and report them to relevant authorities while utilizing machine learning techniques to enhance prediction accuracy. Despite the availability of smart technologies like IoT, machine learning, and artificial intelligence, there is a lack of effective techniques for detecting and preventing road anomalies such as potholes. The model also utilized GPUs to accelerate deep learning processes, albeit at the cost of significant power and energy consumption. Additionally, data from vibration and GPS sensors were utilized to evaluate road surface quality. By leveraging image processing techniques, the system successfully detected potholes from input images, enabling effective identification of such road hazards.
ISO 9001:2008 Certified Journal
|
Page 1615