Hardware Acceleration of Visual Object Search

Open Access
Okafor, Ikenna J
Graduate Program:
Computer Science and Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
April 05, 2017
Committee Members:
  • Vijaykrishnan Narayanan, Thesis Advisor
  • Computer Vision
  • Sliding Window
  • ROI
  • Hardware Acceleration
  • FPGA
Visual Object Search, the process of locating an object within an image, is a key task in many automated vision systems with applications ranging from surveillance to medical imaging. The task is typically performed using one of two methods: an exhaustive/semi – exhaustive search, or region proposal followed by classification. In practice, reasonable classification accuracies, especially in real time systems, have been achieved by incorporating the latter to avoid expensive searching of the entire scene. However, localizing the object within the scene still presents a challenge to visual object search systems. Hardware acceleration has the potential to remove this dependency by making exhaustive/semi-exhaustive image search feasible from a latency perspective. This work aims to investigate the computational performance benefits of using either method, provided the opportunity for hardware acceleration.