Abstract
Detecting passengers and tracking them in public transport is a crucial task in the reality of Smart City(SC) environment. This task combines with additional calculations like passenger counting and utilizes technologies in Computer Vision(CV) field. We explore and evaluate FairMOT hybrid model for passenger counting. It is a state-of-the-art multi-object tracking system that integrates detection and Re-Identification(re-ID) within a single network. This model was trained on 180 videos with a total of 70,710 frames of open source videos containing passenger flow in public buses. Evaluation on different videos showed maximum Recall(97.9%) and MOTA(82.5%) scores. Also moderate IDF1(73.6%) scores. It shows great tracking accuracy and introduces a computationally efficient framework, offering transit operators a scalable and effective solution to optimize passenger flow analysis in SC environment.