Damaged road detection using Image Processing and Deep Learning

Abstract

Computer Science has evolved enormously in the last few decades. It has now far exceeded the Human and Computer interfaces. Its recent sights are scaling, measuring, object detection, etc. Image processing and deep learning have gone through many groundworks in the last few years. Our research paper, based on YOLO V4, LeNet-5, Retina Net, and Faster-RCNN algorithms, proves that these algorithms can detect damaged roads and analyze whether we can enhance any new ways to improve the damaged road detection in real-time. However, in a real-world scenario, it is essential to comprehend the various damages in taking the appropriate action. Thus automotive industries are looking forward to innovations that can increase the efficiency in damage categorization. Along with worldwide industrialization, road damage detection systems have become significantly important both in terms of maintenance and establishment. As the Artificial Intelligence sector is making a lot of progress, faulty road detection through Image Processing and Machine Learning has proved to be a flourishing technique. We can detect damaged roads within specific provisional categories with combinations of such technological stems. In our solution, we propose a futuristic deep learning method for object recognition with the help of four different algo rithms. More specifically, our approach uses a convolutional neural network to train our model with a large dataset solely made for the project and categorize the results into a set of damages along with its comparative analysis.