A deep approach for an adaptive cruise control system

This is a computer vision project for the development of an ADAS system of an adaptive cruise control system. Below it is possible to see some results of our pipeline

Installation

It is strongly suggested to create a virtual enviroment before installing. After venv creation, you can install all the requirements by launching on terminal this command:

pip install -r requirements.txt

Test

In order to correctly test the network you have to download the 2 network weights (YOLACT and GTSRB). You can download them from this Google Drive folder.

• put STNgtsr_55_8736.pth file under GTSRB folder in ./GTSRB/weights folder of this project
• put yolact_base_14_5565.pth file under YOLACT folder in ./weights folder of this project

Once that, you can follow these command to execute the complete network on an offline video (you can choose one of ours videos from this link, pay attention to the HQ and LQ folder):

• If you want to evaluate high_quality videos you must specify --resolution=high_quality

python eval.py --trained_model=weights/yolact_base_14_5565.pth --gtsrb=GTSRB/weights/STNgtsr_55_8736.pth --top_k=15 
--video_multiframe=4 --video=inputVideoPath.mp4:outputVideoPath.mp4 --distance=yes --resolution=high_quality

• If you want to evaluate low_quality videos you must specify --resolution=low_quality

python eval.py --trained_model=weights/yolact_base_14_5565.pth --gtsrb=GTSRB/weights/STNgtsr_55_8736.pth --top_k=15 
--video_multiframe=4 --video=inputVideo.mp4:outputVideo.mp4 --distance=yes --resolution=low_quality

If you want to test this pipeline in your own videos you have to adjust the camera parameters before launching the test.

About the project

This project is based on YOLACT [1], a real-time instance segmentation network. It is the starting point of our pipeline, which can be seen in the image below:

more details are explained in this pdf paper. Above are illustrated some other test examples:

Contributors

This project has been developed by:

• Alessandro Crescenzi - alessandrocrescenzi@outlook.com
• Giovanni Lambertini
• Alvaro Gjepali

References

[1] Daniel Bolya, Chong Zhou, Fanyi Xiao, and Yong Jae Lee. Yolact++: Better real-time instance segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020.