Improving Road Maintenance and Safety through Weighted Ensemble of Deep Convolutional Neural Networks: Focus on Pothole Detection

Road surface irregularities influence an extensive range of aspects-from the general road condition, from the general safety of drivers and mechanical operation of the vehicles up to fuel consumption. If these anomalies exist, they will make driving dangerous, with accidents happening, more damage to your vehicle, and more fuel expenses because of ineffective driving conditions. All these issues have led to various approaches being developed for automatically monitoring the state of the road surface. These methods should help receive accurate data on the smoothness degree of the road and about the potholes' locations. One such unusual approach, which is part of crowdsensing, concerns road surface monitoring with accelerometers built into smartphones. It exploits the ever-growing use of smartphones by using the inbuilt sensors in them to collect data about road conditions as people travel and then analyse this same to provide maps of areas that have significant anomalies in road surfaces. It provides essential data for the attempts being made for maintaining and developing the road. Many advantages of the crowdsensing method are known by now. It is very cheap and, therefore, widely accessible. However, it suffers from false positives caused by structures created by humans, behaviors of drivers, and regular road surface elements that are not true anomalies. Throughout our study, we proposed an ensemble deep-learning approach for pothole discovery classification. Because the model proposed in the present study has an accuracy rate reaching 97.49%, our experimental results prove that this model is better than other currently used models.