Personal tools

Accelerator for Boosted Binary Features

From iis-projects

Jump to: navigation, search
Current feature extraction system.

Short Description

Image feature extraction is an important analysis tool in many computer vision applications. In the context of this project, it is specifically used for sparse depth estimation in stereo video, and sparse flow estimation in normal video. This project builds upon a previous Semester Thesis (Feature Extraction with Binarized Descriptors: ASIC Implementation and FPGA Environment), where a hardware architecture for image feature extraction has been developed. This IP is currently used in the video analysis part of a more complex video processing system which performs automatic multiview conversion in real-time.

The goal of this project is now to extend the existing hardware architecture with a new type of descriptor which has recently been developed at the IIS. This new descriptor uses a different set of low-level features which have been trained using the AdaBoost learning algorithm, and compared to other standard algorithms it shows much better performance - at relatively low computational cost.

Status: Completed

Scope: Semester or Master Thesis
Looking for 1-2 Interested Students
Supervisors: Michael Schaffner, Pascal Hager

Prerequisites

VLSI I
Introductory course in computer vision (recommended)
Interest in computer graphics / computer vision
Matlab, VHDL and C++

Character

25% Theory & Literature Study
35% Matlab Evaluations
40% Hw Architecture & FPGA Implementation

Professor

Luca Benini

Partners

Disney Research Zurich

↑ top

Detailed Task Description

Goals

Practical Details


↑ top