<![CDATA[This paper examines the potential and reliability of AI-based video analytics for solving key traffic engineering problems. The objectives were to compare several commercially available tools for collecting traffic data and, through practical examples, to show that AI-processed data can be used for the development, calibration, and validation of traffic models. Four AI-based video analytics (StreetLogic Pro, DataFromSky, CVEDIA RT Studio, and Camlytics Single) were tested using field video recordings at a signalized intersection on an urban arterial in Split, Croatia. Detection accuracy, usability, and sensitivity to camera placement and recording conditions are analyzed, and selected microscopic parameters (saturation flow rate and control delay) were obtained and compared with values derived from HCM procedures. DataFromSky and CVEDIA RT Studio achieved 97–99% vehicle detection accuracy and provided detailed trajectory data suitable for scientific applications, while StreetLogic Pro achieved 100% accuracy for operational vehicle counting. AI-based estimates of saturation flow rate and control delay differed by less than 1% and 5%, respectively, from traditional field measurements. The main novelty of this research lies in its practical comparison of AI-based video analytics tools combined with a worked example of using AI-derived data to calibrate analytical models, providing practical guidance for researchers and practitioners in traffic engineering.]]>
