Publication Date: 2023/08/18
Abstract: Potholes are one the major concerns of underdeveloped and developed nations. Roads are the essential means of transportation for a country and when this road becomes bad safe driving is threatened, this may result to road traffic crashes hence the need to provide an intelligence system that can detect this potholes in real time and give drivers real time feedback to enable them make adequate decisions while driving. This paper presented the pothole detection model that was trained with data extracted from Google and some real time data. Region Convolution Neutral Network (R-CNN) as an object detection model was used to analyze images captured via cameras used for image detection specifically for Pothole Detection, and only part of the image is processed instead of the background; hence very large data and consequently tedious computations, pixel matching, parameter updating and sorting were significantly decreased. This work used the comparative analysis, Microsoft Common Object in Context (COCO) and TFLITE mobile net. The model was evaluated and their strengths and limitations were analyzed based on metric parameters such as accuracy, precision and F1 score. The results analyzed show that the suitability of the algorithms over is depended to a great extent to the use cases they were applied in. In a good testing environment, Region Convolution Neural Network (R-CNN) gave a good classification report with the parallel testing proof that the model is not perverse.
Keywords: Machine Learning, Object detection, Pothole detection, Region Convolution Neural Network.
DOI: https://doi.org/10.5281/zenodo.8262321
PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT23JUL1901.pdf
REFERENCES