DEVELOPMENT OF EDGE CLOUD COLLABORATION TO INCREASE ACCURACY AND REDUCE LATENCY IN ROAD DAMAGE DETECTION

• Main Author: Andika, Furqon
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81153
• Institution: Institut Teknologi Bandung

One important aspect of road maintenance is recognizing the type of road damage. Various studies have been carried out to identify various forms of road damage automatically. Automatic identification is generally carried out on the server, but this method has weakness due the latency. In this research, we utilize edge-cloud collaboration to detect road damage. The dataset used is RDD2022. This dataset will be trained with several algorithms, then we compare the accuracy results to determine the best algorithm. We also carry out a preprocessing stage with image enhancement to increase detection accuracy. This image enhancement process is used to balance the dataset and add more features to an image. This image enhancement process increases accuracy by 1.2%. Then, we apply the selected model to be applied to edge devices to detect road damage. Then images that are not successfully detected will be sent to the cloud to be stored and re-trained. The model from this training will later be used to update the existing model on the previous edge. Implementation of this edge-cloud collaboration system results in faster detection time with a detection accuracy of 88.25%.