This paper presents a comprehensive methodology for constructing an annotated traffic database designed for vehicle detection using the YOLO-NAS model. The dataset was generated from real-time video footage captured at regular intervals from a traffic camera in Coldwater, Michigan, USA. Over one week, a total of 1,155 images were collected during daylight hours and manually annotated with bounding box coordinates and class labels for vehicles, including cars, buses, trucks, and trailers. Each image is accompanied by timestamped vehicle counts, offering valuable temporal insights into traffic patterns. The methodology outlines the data collection, image annotation, and database organization processes, providing a replicable framework for generating traffic datasets. The resulting dataset is intended for future use in vehicle detection and traffic analysis, with potential applications in urban planning, traffic management, and smart city development. Additionally, the effectiveness of the YOLO-NAS model in detecting different types of vehicles in an uncontrolled environment was evaluated, demonstrating its robustness and applicability across varied traffic conditions. This scalable approach can be applied to various urban environments, making it a valuable contribution to traffic data collection and machine learning-based traffic analysis.