Abstract
Modern outdoor self-localizing computer vision applications require descriptors more than repeatability. The descriptors need to be invariant to light conditions and transformation changes to give support for efficient classification. This paper investigates a new framework based on the genetic algorithm to create and optimize extensible modular descriptors for specific outdoor environments. The algorithm returns descriptors with improved efficiency and classification performance. It controls the image processing and machine learning parameters and optimizes the descriptor size by activating the necessary modules. To show the strength of the descriptor, we compared it with the most commonly used standard descriptors on speed, accuracy, and invariance to light conditions, image resolution changes, scale, affine transformation, rotation, and classification. The results show that it has an average result in transformation invariance, and its description ability in sparse areas is significantly better than that of the most-used descriptors. The descriptor was also integrated into an augmented reality algorithm to create a self-regulating segmentation application.
Highlights
When we perform localization or segmentation tasks in computer vision using visual input, it is essential to be able to compare the descriptions of the visual features taken from the incoming images
The primary approach is to create and train light modular descriptors composed of simple mathematical calculations that are chosen for the selected environment, with machine learning techniques using EDD [1] as a base
The results of each genome’s random forest were turned into a confusion matrix. This served for the precision calculation and comparison and for measuring the patch classification correctness by the overall recognition rate and the average recognition rate of each feature descriptor defined by the genomes
Summary
When we perform localization or segmentation tasks in computer vision using visual input, it is essential to be able to compare the descriptions of the visual features taken from the incoming images. A different method is to use a light image-processing element for invariant local features such as image moments [12], which creates a scale- and rotation-invariant descriptor but is only applicable in constrained environments in vehicle detection scenarios. The above works aim to create descriptors that function in a broad spectrum of areas As this is hard to achieve, different techniques were introduced to extend or optimize existing local features to make them invariant across environmental changes. The retrieved local descriptors have to consist of reproducible, invariant and unique features, but the calculation process might vary widely, depending on (1) the computation technique used (e.g., histogram- or gradientbased), (2) the component data type (e.g., floating-point or binary), and (3) the application environment (e.g., indoors or outdoors). The calculated attributes cannot be affected by environmental alterations (scale, rotation, transformation and light changes) and have to be compact enough for near-realtime performance
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
