Automatic Passenger Counting (APC) is a key element of Intelligent Transportation Systems (ITS), which can aid in monitoring of passenger occupancy and public transport management. But in the case of bus counting it is still difficult to count the number of passengers, because of overlap, partial view and the camera's point of view. This paper proposes a Pose-Based Keypoint-Triangle Framework for passenger counting (PKTF) by integrating YOLO-based person detection, human pose estimation, 2D nose-shoulder keypoint validation, multi-object tracking and temporal smoothing: the framework is complemented with the design of privacy-preserving metadata communication. Different YOLO detection and pose-estimation networks were used to test the framework and counting performance was evaluated by the mean absolute error (MAE). YOLO11s achieved the best performance with an MAE of 1.38 persons per frame while the YOLOv8n and YOLO11n models achieved an MAE of 2.58 and 2.38 persons per frame respectively. But the MAE of YOLO11n-Pose increased to 3.96 persons per frame if the first keypoint-triangle validation thresholds were applied, which indicates the need to improve the geometric constraints. Another ablation study discovered that median smoothing modestly reduced the counting error of the detection baseline (2.38 to 2.21) but did not change that of the validated pose pipeline, while tracking increased the counting error in both configurations. Computational evaluation under a CPU-based configuration showed processing speeds ranging from 4.3 to 11.5 FPS. The results show that it is vital to consider the geometric validation and temporal processing separately when designing vision-based passenger-counting systems. The framework needs some further development of the geometric constraints and detailed evaluation in various passenger and camera settings to enhance its reliability.

Keywords—automatic passenger counting, YOLO, human pose estimation, keypoint-triangle geometry, multi-object tracking, temporal smoothing, computer vision