Vignette-Car-Vision

Vignette on identifying objects such as pedestrians, traffic lights, and other road-side objects using computer vision. This was a project for PSTAT197A.

Contributors

Sanil Katula, Bahaar Ahuja, Shenyi Jiang, & Calder Glass

Abstract

Self-driving cars need to be able to identify common entities on or alongside the road. The dataset "Udacity Self-Driving Car" on Roboflow provided 15000 images of pedestrians, cars, traffic lights, bikers, etc and labels containing the coordinates for those objects. The YOLOv7-PyTorch model classifies the different potential roadside objects under 11 labels and assigns a probability that the object is of that given label.

Why use YOLOv7?

The YOLOv7 model was chosen as it's better at precision of small objects such as traffic lights and pedestrians, when the objects are either a long distance away or captured with a low-resolution camera. The model is also more robust when dealing with day/night lighting, weather, and information-dense images. Additionally, implementing it via PyTorch allowed for the model to be easier to fine-tune to the given traffic dataset in this vignette.

Repository Contents

• data/ - Contains the general structure of the dataset when it's downloaded manually from Roboflow. The images and labels were not included because they would exceed GitHub's file limit.

  • export/ - Replicates the file structure that is used to path the images for training, validation, and testing.
    • images/ - Contains the images of road-side objects that the model is trained on.
    • labels/ - Contains the labels for each given image, where the integer is represents the given classification of an object in that image.
  • README.dataset.txt - Gives a brief overview of the dataset and its differences from the original version on GitHub.
  • README.roboflow.txt - Mentions that the dataset was exported from Roboflow, the number of images in the dataset, and the preprocessing applied to the images.
  • data.yaml - Denotes that training and validation subfolders of the images folder and the 11 classes of objects that can appear in the images.

• draft/ - Contains the rough draft of the Jupyter Notebook used for the project.

  • 197_car_vision.ipynb - Rough draft of the Jupyter Notebook of the vignette.
  • Jupyter Notebook -- generated with runcell.pdf - Pdf render of the rough draft of the vignette.

• scripts/ - Contains the final script/final draft of the Jupyter Notebook of our code.

  • 197_car_vision_notebook_final_annotations.ipynb - Final draft/script with some line annotations of the Jupyter Notebook used for the vignette.
  • 197_car_vision_final.pdf - Pdf form of the final draft of the vignette.

How to replicate computer vision on a different dataset

1. Clone the YOLOv7 repository
2. Load Pretrained Model
3. Download Dataset
4. Split test/train
5. Sample Labeling on a image

References for Further Learning

Learning Open CV 4 Computer Vision with Python 3: https://books.google.com/books/about/Learning_OpenCV_4_Computer_Vision_with_P.html?id=ef_RDwAAQBAJ&source=kp_book_description

Computer Vision: Algorithms and Applications, 2nd Edition: https://szeliski.org/Book/

Computer Vision, From 3D Perception to 3D Reconstruction and Beyond: D. A. Forsyth and J. Ponce. Computer Vision: A Modern Approach (2nd Edition). Prentice Hall, 2011.

Dataset - Udacity Self-Driving Car on Roboflow

https://public.roboflow.com/object-detection/self-driving-car