Local Invariant Feature Detectors: A Survey

Abstract

In this survey, we give an overview of invariant interest point detectors, how they evolved over time, how they work, and what their respective strengths and weaknesses are. We begin with defining the properties of the ideal local feature detector. This is followed by an overview of the literature over the past four decades organized in different categories of feature extraction methods. We then provide a more detailed analysis of a selection of methods which had a particularly significant impact on the research field. We conclude with a summary and promising future research directions.

Hollywood 3D: What are the Best 3D Features for Action Recognition?

“…The extended algorithms discussed in this section are based on the Harris Corners work by Laptev and Lindeberg (2003), the Hessian points algorithm by Willems et al (2008) and the Separable Filters technique by Dollar et al (2005). For a comparison of the original 2D interest point detection schemes (without the proposed depth-aware extensions), see the survey paper by Tuytelaars and Mikolajczyk (2008).…”

An Assistive Vision System for the Blind That Helps Find Lost Things

“…In their system, computer vision makes it possible to identify and locate objects such as signs and landmarks. To this end, they rely on the Scale-Invariant Feature Transform (SIFT) by D. Lowe (see [10]). However, the corresponding evaluation has been performed in a simplified scenario and computer vision was left as major aspect for future work.…”

“…However, the corresponding evaluation has been performed in a simplified scenario and computer vision was left as major aspect for future work. Bigham et al [2] use Speeded Up Robust Features (SURF; see [10]) for object identification, but instead of training an object database (see, e.g., [3]), they send images with user requests (e.g., where is the object in the image) to Amazon's Mechanical Turk [1] where humans can outline the objects. The outlines of the object can then be used to estimate the object's location in the environment and guide the user towards the object by informing the user how close he is to the target [2].…”

“…Specific Objects: We use SIFT (see [10]) to detect known objects. To this end, our system provides a simple training interface, which makes it possible to train new objects by holding them in front of the camera and triggering snapshots.…”

A Finite Element Blob Detector for Robust Features

“…To deal with the many natural scales at which features may be present, corner detectors have been developed to work on multiple scales, thereby having the ability to detect all corners. Many of these corner detectors do not only detect actual corner points but also other "interesting points" that may not strictly be recognized as corners [8,13]. For some particular applications the ability to detect interesting points that are robust to changes within the image is seen as a more desirable characteristic than specifically detection of real corner points.…”