Queued Commuter Counting System by using Caffe Deep Learning Technique

Abstract

The problem of using public transport is that the volume of transport vehicles is insufficient for the number of commuters. Commuters who have to wait in line lose time and opportunity. Solving this problem by increasing vehicle estimates might not be easy due to budget constraints. Transportation management is essential in order to meet the needs of commuters as much as possible. In this article, we proposed a system for displaying the number of commuters queuing at tram stations using a visual identification technique. The system consists of Raspberry Pi 3, Web camera, Firebase application, Caffe model, personal computer and HTML. A key component of the system is the person identification by the Caffe deep learning technique. Commuters pictures at each station are captured and processed. Finally, the quantity of the commuters are displayed on a website. We tested the system in a case study of the electric cars transportation on campus. The experimental results showed that the system worked well with the nearest distance of the passenger to the camera. The lower number of passengers, the better accuracy of person detection (different testing period). The result of detection people without mask on showed 100% accuracy for all distances.