Logo image
Implementation of real-time traffic light and color detection using FPGA-based acceleration
Thesis   Open access

Implementation of real-time traffic light and color detection using FPGA-based acceleration

Harish Venkatesh Babu and Tarun Karanam
California State University, Sacramento
Master of Science (MS), California State University, Sacramento
08/21/2025
Handle:
https://hdl.handle.net/20.500.12741/rep:13370

Abstract

This project uses the Ultra96 FPGA board to detect real-time traffic signals and recognizeobjects with the help of hardware acceleration and Python-based algorithms. It showcases the FPGA's processing ability by handling inputs from two USB cameras. A Logitech USB camera and a monochrome USB camera. At the same time. The implementation incorporates sophisticated detection algorithms such as TensorFlow Lite, YOLO, and Haar Cascade, resulting in speed and accuracy [39]. The system for detecting is based on a TensorFlow Lite model that has been trained and optimized for edge devices using methods like quantization. The training data consisted of 1,000 images of traffic lights taken under various lighting and weather conditions. To improve the model's robustness, data augmentation techniques like rotation, scale modifications, and brightness alterations were used. The system identifies traffic signals as objects. Understands their color (red or green), allowing for accurate on-the-spot detection. This project highlights the advantages of FPGA hardware acceleration over processors such as Raspberry Pi. The Ultra96 FPGA demonstrates superiority in speed and efficiency while offering scalability suitable for utilization in intelligent traffic control systems, self-driving vehicles such as autonomous cars, and roadway safety surveillance. System deployment is performed via Jupyter Notebook to provide an adaptable development workflow. The images capture the system's strengths well. The Logitech camera succeeds in displaying resolution and correctly detecting colors, whereas the monochrome camera does well in low-light conditions. This clever blend of FPGA acceleration with a software framework and real-time object detection provides a flexible and feasible answer for advanced traffic control systems.
pdf
KaranamT_VenkateshbabuH_Fall2024_pdf_508CompliantCopy_-_Copy5.01 MBDownloadView
TextProject Open Access

Metrics

4 File views/ downloads
10 Record Views

Details

Logo image