vraj141's repositories
IMAGE-PROCESSING-FOR-INDIAN-ROADS
In developing nations such as India, the vehicular growth rate is increasing exponentially which is worsening the traffic operations. Most of the urban cities in India are facing traffic related problems such as congestion, accidents, pollution, etc. during peak hours. The main cause for traffic congestion in such cities is mainly due to uncontrolled urbanization and extensive usage of private vehicles. The traffic congestion leads to many problems like increase in travelling time, health disorders and accidents. Road accidents in India claimed over 1.5 lakh lives in the country in the year 2018, with over-speeding of vehicles being the major cause. The Ministry of Road Transport and Highways report on Road accidents in India stated that road accidents increased by a rate of 0.46 % in the year 2018 when compared to 2017 . Due to this there is a need to develop a model which can analyze and detect poor road conditions like potholes.This project aims in building a system which can detect the poor road conditions and can notify the driver as well as the government beforehand to improvise the road conditions. Pothole detection is being carried out using two techniques namely image processing and machine learning techniques. Those two techniques are used for a study of the detection and occurrence of potholes. In this project, we implemented both of them individually and then a combination of the techniques to see how image pre-processing can affect the performance of a deep learning model. The image pre-processing steps like erosion,median blur etc applied in this project removes the noise in the image which helps in better training of the model.First of all, we implemented the image processing techniques on a single image in the order: median blur, erosion, canny edge detection, contour detection, bounding box prediction. After that we labelled a dataset of around 800 images and passed it to the YOLOv5 model and noted the results.Secondly, we applied median blur on the already labelled dataset and then passed it to the YOLOv5 model and noted the results. Lastly, we applied median blur and erosion both and passed it to the model and noted the results. We compared the results at last. The conclusions reached are that a combination of machine learning and image processing techniques generates good performance in pothole detection and machine learning techniques provide better results than the usual image processing models.