Vehicle Detection based on Faster R-CNN

This project is a vehicle detection application for vision-based Intelligent Transportation Systems (ITS). These systems utilize roadway camera outputs to apply video processing techniques and extract the desired information, which is vehicles' visual features in this special case.

Algorithms/Architectures

1. Faster R-CNN: Since deep learning has become a popular research subject of machine learning and attracted a huge attention compared to the traditional approaches, we have utilized Deep Neural Network (DNN) in this project. Among different DNN and CNN approaches, Faster R-CNN is another approach similar to Fast CNN in which the region proposals are extracted by the means of a separate network instead of utilizing the selective search module (read more)
2. ResNet-50: is a 50 layer residual network and the main purpose of utilizing this framework is to improve the classification results, besides preventing accuracy degradation in training deeper layers.

Inputs/Outputs of the System

The input of the system can be a single or set of vehicle images for training/testing purposes. The ouput of the system is the same image(s) with visual bounding-boxes to cover vehicle inside the image. Sample result can be seen in below image:

Environment

The project is implemented by MATLAB 2018 with the aid of its deep learning libraries. There are several standard datasets to train/test the project like the Cars Dataset by Stanford Artificial Intelligence Library (link) and other standard real condition vehicle datasets.

Publications and CopyRight

We have presented the architecture, experiments and calibration settings in the paper below:

@inproceedings{Tourani2019,
	author = {Tourani, Ali and Soroori, Sajjad and Shahbahrami, Asadollah and Khazaee, Saeed and Akoushideh, Alireza},
	title = {{A Robust Vehicle Detection Approach based on Faster R-CNN Algorithm}},
	booktitle = {4th International Conference on Pattern Recognition and Image Analysis, IPRIA 2019},
	doi = {10.1109/PRIA.2019.8785988},
	isbn = {9781728116211},
	pages = {119--123},
	year = {2019}
}