Vehicle Type Classification and Detection System using YOLOv7-tiny Model on Single-Board Computer

Abstract

Transportation is playing an important role for human civilization, for example transportations is being used as distributing goods and products. Therefore, the total numbers of vehicles as a part of transportation will continue to increase every year. But in Indonesia, the majority of its people is still using their personal transport rather than public transportation. This is supported by the data of total number of vehicles in Indonesia from 2018 – 2020, which is shows that personal transport is still dominant than public transportation. The causes of traffic jams is a result of various factors, such as the roads are not designed to accommodate the increasing number of vehicles, insufficient traffic signs, and poor traffic management. The road traffic data is one of the aspects that could reduce traffic jams. The process of collecting road traffic data which is still done manually has several shortcomings, such as it takes a long time and there may be errors due to human error. This research has a goal to create a vehicle type detection and classification system that have a good detection accuracy and detection speed that can be run on single-board computer devices. YOLOv7-tiny model that performs detection and classification using input from video on the NVIDIA Jetson Nano device gets a True Positive (TP) score of 96.58%, a False Positive (FP) score of 0.98%, and a False Negative (FN) score of 2.44%. YOLOv7-tiny on the NVIDIA Jetson Nano device can run with an average Frame per Second (FPS) of 6 FPS.

GS Cited Analysis