<p>For the safety of vehicles and the welfare of users, a well-kept road network is a crucial prerequisite. One of the key measures of the quality of a road is its surface condition. Usage of public cameras can be an inexpensive solution to implement a real-time road surface surveillance and alerting system. This work uses an improved YoloV3 for quality detection. The proposed method pinpoints the exact location of the road surface damage by storing the coordinates, such as latitude and longitude, on the server. The alert can be used to allocate resources right away in an emergency. This study analyses deep transfer learning's potential to resolve the multi-class identification problem of differentiating moist surfaces from water. Two transfer learning methods were used to retrain the pre-trained deep convolution neural networks: a pair of techniques is feature extraction and fine-tuning. YoloV3, YoloV4-tiny, and YoloV5 are three cutting-edge pre-trained models trained on a dataset of mobile-captured photos of structured and unstructured roads, speed bumps, wet surfaces, and potholes with and without water. These findings demonstrate that improved YoloV3 surpassed the other two models with a classification accuracy of 98.69%. Additionally, it was shown that optimizing the previously trained models increased the detection accuracy.</p>