Citation

Sooraj PSA, Kollerathu V, Sudhakaran V. J. Big Data Anal. Transp. 2021; 3(2): 109-118.

Copyright

(Copyright © 2021, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s42421-021-00044-1

PMID

unavailable

Abstract

Automatic traffic counting and classification (ATCC) is a salient step in many applications such as accessing the contribution of traffic to air pollution for clean air strategies and computing the passenger car unit (PCU) for urban road infrastructure planning and management. This work focuses on developing an ATCC system that is low cost, privacy-preserving, and auditable using state-of-the-art AI technology on mobile phones. The camera unit and the GPU compute available within a mobile phone are used to capture the video feed and run the required analytics for detection, tracking and counting in real time. On the target device, we have been able to achieve 12 FPS. On the test data composed of four videos, the solution achieved a counting precision and recall of 0.96 ± 0.02 and 0.86 ± 0.03, respectively.

Language: en

Keywords

Clean air strategies; Deep learning on mobile devices; MobileNetSSDLite V3; Traffic counter; Vehicle detection