Undergrad Research Project - SMART Headlight Hardware Acceleration

Spring 2016

Brandon Perez
James Hoe
Project description

This project is in conjunction with the SMART headlight team, part of Carnegie Mellon's Illumination and Imaging Lab (ILIM). The overall goal of the SMART headlight team is to improve safety in low light and poor weather conditions by leveraging programmable headlights. The goal of this project is to enable calculations in real-time by using a Xilinx Zynq FPGA system to accelerate the computer vision tasks required to evaluate the environment.

The project currently has a simple pipeline that performs homography, to align the camera frame with the projector's (headlight's) frame. First, we will port the 18-545 capstone project implemented this system to the project's board. This will use Harris corner features to track regions of interest in the camera. Next, we will use a blob detector to track headlights in the image. We will use these detections to identify vehicles in the scene. With the vehicles identified, we will programmable the headlight's output so it does not shine into the opposing driver's windshield, reducing the glare. We will be using Xilinx's Vivado tool to program the FPGA, and will exploit the full SoC architecture of the Zynq processing system to develop both hardware and software to achieve our goal. We expect to have a blob detector running in real-time by the end of the semester, able to track other cars in the scene.

Return to project list