A deep learning pipeline and an image-scraper to count pedestrians from traffic camera images on 511NY. The utilized deep learning model is the Faster R-CNN built upon the ResNet-50 architecture, pre-trained on MS-COCO dataset. Other deep learning models can easily be swapped in with the utilized GluonCV library. This work represents a proof-of-concept, and higher accuracy can be achieved through more rigorous model training and a more specialized, labeled training dataset containing more similar images with labeled pedestrians.
- E. Deleu, S. Elez, A. Gadodia, K. Macvaugh and G. Zhao, "Using Deep Learning for Urban Pedestrian Counting," 2021 IEEE MIT Undergraduate Research Technology Conference (URTC), Cambridge, MA, USA, 2021, pp. 1-5, doi: 10.1109/URTC54388.2021.9701649.
@INPROCEEDINGS{deleu2021pedestrians,
author={Deleu, Edward and Elez, Stefan and Gadodia, Ansh and Macvaugh, Kyra and Zhao, Grace},
booktitle={2021 IEEE MIT Undergraduate Research Technology Conference (URTC)},
title={Using Deep Learning for Urban Pedestrian Counting},
year={2021},
volume={},
number={},
pages={1-5},
doi={10.1109/URTC54388.2021.9701649}
}