A-study-of-nighttime-vehicle-detection-algorithms

This is my master thesis. Whole code is messy but right now but it could be useful for somebody.

Link to dataset: https://doi.org/10.34808/e8ht-d443 https://mostwiedzy.pl/pl/open-research-data/a-study-of-nighttime-vehicle-detection-algorithms,10070936421042877-0

Link to thesis: http://download.panjacob.pl/Badanie_algorytm%c3%b3w_wykrywania_pojazd%c3%b3w_w_porze_nocnej.pdf

The study used 4 methods to identify vehicles:

• Detection on image

• Classification of detection results on image

• Image classification

• Classification of light intensity samples

  

Testing platform

Nvidia Jetson Nano, Raspberry Pi 4 (both 4GB) and Google Coral USB Accelerator.

Training

Optimalization

Performance of networks

The work did not take into account the braking distance of the vehicle. With breaking distance it is required that algorithms performs inference at 2.2 FPS.