Publication Date: 2022/04/27
Abstract: Object Detection Potholes are a traffic hazard, endangering the safety of both automobiles and pedestrians. It is one of the leading causes of road accidents and the loss of lives and property in most developing countries. As a response, there is a need to collect and update data on current road conditions on a regular basis so that vehicles may be warned of other routes and the appropriate government department can take urgent action to remove potholes for the benefit of commuters. Using object identification algorithms on photos captured with a smartphone camera is a simple and effective technique to locate potholes on roadways. As a result, the goal of this research is to evaluate the performance of state-of-the-art neural network algorithms such as YOLO and Faster R-CNN with VGG16 and ResNet-18 architectures for rapid and accurate pothole identification. Furthermore, an updated YOLOv2 architecture is suggested to address the "pothole" and "regular road" class imbalance problem, and its performance is compared to that of existing object recognition algorithms utilising accuracy, recall, intersection over union, and frames processed per second (FPS). For real-time geotagged pothole recognition from images, this model can be used in autonomous cars. Pothole detecting software may also offer alternative environmentally friendly routes and assist commuters with low-light navigation.
Keywords: Autonomous Vehicle; Deep Learning Neural Network; CNN; YOLO Algoritham Object Detection; Image Processing .
DOI: https://doi.org/10.5281/zenodo.6496700
PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT22APR029_(1).pdf
REFERENCES