Robust System for Counting Movement-Specific Vehicle at Crowded Intersections in HCM City

Le-Hoai-Hieu Dang, Van-Tin Luu, Vu-Hoang Tran, Ton-Nghia Huynh

Abstract: Traffic jam is a difficult problem in many countries around the world, especially Asian countries. In addition to messing up human timetables, it also affects people’s physical and mental health because of waiting for a long time under an atmosphere full of dust and pollution. Traffic congestion will also increase local emissions that have a negative impact on the environment. Therefore, more and more researches are being done on counting, tracking, and predicting vehicle traffic on the road. However, traditional traffic management systems focus on counting total vehicles in one frame or in consecutive frames regardless of the direction of movement. This makes it difficult to predict traffic situations in neighboring areas in the near future. Besides, Asian environments, where motorbikes are used with high density making it difficult to count, have yet to be paid much attention. In this paper, we proposed a robust vehicle flow counting system with movements of interest (MOI) which can work in complex environments like Asian countries. In other words, we will detect and track objects in the region of interest (ROI) then count traffic in different directions. The experimental results conducted on HCM AI-Challenge 2020 dataset show that our system can work effectively in almost any environment.