A portion of the country's road network is equipped with surveillance cameras. These cameras are continuously monitored by human operators. Given the large number of cameras, this approach to traffic monitoring is not only labor-intensive and costly but also highly error-prone. Therefore, an automated framework with real-time processing capabilities is needed to extract vehicle trajectories that transportation researchers and practitioners can use to investigate microscopic vehicle behavior in various contexts. This study aims to design an advanced and efficient framework for extracting accurate and continuous vehicle trajectories using real-time object detection and tracking in video-analytics-based surveillance systems. In this study, the YOLOv10 algorithm was employed for object detection, while ByteTrack was used for object tracking. Recognizing the importance of access to a diverse, high-quality, and representative dataset, a dataset named Surveillance Cameras was developed from a variety of traffic scenes with the aim of providing high-quality annotations. Transfer learning was employed to fine-tune the pre-trained YOLOv10m model on this dataset. The fine-tuned model achieved precision, recall, and mean average precision values of 96.3%, 95.2%, and 98.3%, respectively, with an inference latency of 5.08 ms. The results demonstrate that the proposed framework can accurately detect and correctly classify most vehicles while ensuring a high frame rate in real-time applications. Within the tracking-by-detection paradigm, this performance improves the data association process and, while maintaining track-ID continuity across consecutive frames, enables the extraction of accurate, stable, and continuous vehicle trajectories. The proposed framework provides precise information on the positional changes associated with each track ID over time and can be applied to motion-pattern analysis, position prediction, traffic parameter extraction, and the identification of potential traffic conflicts.