Initial Estimation Passenger Number using YOLO-NAS

The Automation Passenger Counting (APC) system is a key point in transportation, especially in public transportation. It can help the company and passengers estimate the capacity of public transportation and estimate travel times. APC was developed in various ways, including utilizing deep learning. In this study, we do initial research to estimate the number of passengers based on people detection using YOLO-NAS. We compare three types of architecture of YOLO-NAS and different confidence scores. Medium architecture gave the best results, with an average accuracy of 95% and 92.226% for the rear and front views, respectively. Meanwhile, high confidence scores can detect smaller objects but increase false detection. Otherwise, small confidence scores lower false detection but fail to detect small objects and give low estimation numbers. A confidence score of 0.7 has the highest accuracy but has the smallest number of detected people.