A Survey of the Optimization System Road Damage Detection on CNN Model for Edge Device

Road damage survey problems are carried out conventionally. It is necessary to develop tools for automatically detecting road damage using technology and artificial intelligence (AI). The difficulty in detecting road damage is that it necessitates the use of an edge device to examine visual data while collecting road defect data in motion. However, quick detection of road damage objects is crucial because of edge devices' constrained processing and storage capacity. Due to the limited computation and storage capabilities of edge sensors, a road damage detection method with high accuracy and low detection time is required. In this research, we present an overview based on sensor cameras using CNN architectures that can be implemented on embedded targets to increase the performance of road damage detection for edge devices. As a result, it is necessary to develop CNN models of one-stage detectors, two-stage detector and a combination of both models to increase accuracy. Additional study is required to improve system performance in order to boost accuracy and inference time by combining tuning hyperparameter, pruning, and quantization methods.