Pothole Detection System Using Object Detection through Dash Cam Video Feed

In the present work, the possibility of preventing damage to cars and increasing the safety of cars using Image Processing has been explored. A method for modifying an image to produce a better image or to extract some relevant information from it is known as “image processing” [1]. It is a kind of signal processing where the input is an image and the output can either be another image or traits or features related to that image. We have dived deep into Machine Learning and further explored Deep Learning Techniques [2] using the custom trained CNN model. The following paper discusses the possibility of using YOLO Algorithm [3] to detect potholes and alert the driver to slow down thereby reducing possibilities of accidents. Initial Models tested used Yolov3 [4] with a small dataset of 600 images. The final model tested in this paper uses 3000 images of potholes hand clicked at angles ranging from 30 degrees to 90 degrees with respect to the ground. It further dives deep into object tracking along with the YOLOv4 algorithm [5] and implementing the deepSORT [6] algorithm. A little light is also shed on the probability of the use of ML Model to detect the most common roads used at most common timings to keep the driver well informed about all upcoming potholes and road disturbances seen previously.