Low-Cost Object Detection Models for Traffic Control Devices through Domain Adaption of Geographical Regions

Automated inspection systems utilizing computer vision technology are effective in managing traffic control devices (TCDs); however, they face challenges due to the limited availability of training datasets and difficulty generating new datasets. To address this, our study establishes a benchmark for cost-effective model methods that achieve desired accuracy using data from related domains YOLOv5, one-stage object detector known its high speed. In this study, three cases were developed distinct approaches: (1) with COCO-based pre-trained weights, (2) weights source domain, (3) synthesized dataset mixed target domains. Upon comparing these cases, found directly applying domain is unfeasible, small amount necessary optimal performance. A trained fine-tuning-based adaptation minimal data, proved be most resource-efficient approach. These results contribute valuable guidance practitioners aiming develop TCD models enabling them build while conserving resources.