Computer vision for smart library

Matej Trček (2016) Computer vision for smart library. EngD thesis.

Preview

PDF Download (2334Kb)

Abstract

Hands-free interfaces allow us to comfortably integrate technology into mundane tasks. This thesis describes the development of an application for book cover recognition using an RGBD camera for use in a smart library. The application detects a plane within the depth image, finds the corners of a rectangle within it and alligns it with the camera plane. Computer vision techniques are used to compare the recorded image with a prepared database of book covers to find the best match. The depth image is taken with a Microsoft Kinect camera and the application is coded in the C++ language using the OpenCV library. The plane detection uses the RANSAC algorithm while the corner detection algorithm relies on finding the most distant points and checking the corners between them. PCA projection and SIFT feature frequency histograms are used to describe the images classification is performed using SVM and k-NN algorithms and the euclidean distance to the best example of each class.

Actions (login required)