Crowd analysis and management is key area of study for transit agencies to maximize operational efficiency and minimize risk. Passenger flow volume, crowd density and walking speed are key features of crowd analytics. This study proposes a generalized Artificial Intelligence (AI)-based crowd analytics model framework for rail transit stations, by analyzing high-density crowd video data. We propose a generalized triple-layer AI-aided methodological framework (named AI-Crowd) for calculating the flow volume, crowd density and walking speed. For the pedestrian detection and tracking layer, the You Only Look Once (YOLO) and Simple Online and Realtime Tracking with a Deep Association Metric (Deep SORT) are integrated to detect and track the dynamic position of each individual. Then, we adjust the original image from the camera to real scale in the camera calibration layer. Next, the calibrated results used by the metric calculation layer to implement a comprehensive calculation of crowd metrics. Several video records from stair and hallway scenarios in a major rail transit station in China are used to validate the model framework. Based on the example video data samples, the pedestrian counting accuracy can be 95% - 98%; the fundamental diagrams of density speed are shown to be consistent with empirical studies. Furthermore, the methodology has practical applications such as automatic passenger counting, level of service evaluation, and social distancing monitoring in the era of COVID-19.