This project is a class project that I am doing in my CS 469 - Intro to digital image processing class. The topic I chose to make a project on is image semantic segmentation for traffic objects. With the rise of autonomous vehicles, I thought this could be a good project idea and learn how vehicle companies like Zoox handles autonomous driving with their cameras.
This project focuses on developing an advanced algorithm for identifying and separating vehicles, pedestrians, and traffic signs from images through image processing techniques like image segmentation. This process can be very helpful for different software such as autonomous vehicles and even traffic cameras and monitoring systems. With methods of image segmentation, the goal is to accurately identify and mark these different traffic objects.
My approach proposes an approach of using semantic segmentation to identify and separate vehicles, traffic signs, and pedestrians with deep learning. I will implement a semantic segmentation model using pre-trained models like U-Net or DeepLabV3. With this approach, I won't have to build and train my own models, and specifically focus on making the pre-trained model more accurately identify these different common traffic objects.
One publicly available dataset I will utilize is from Cityscapes which provides 5,000 images of urban street scenes. These datasets contain pixel annotations for object segmentation which can be very helpful for my project