Terminus Group's New Breakthrough in Fitting Model Parameters from Outlier-contaminated Datasets

Recently, Terminus Group proposed a method called Latent Semantic Consensus (LSC) to estimate reliable geometric model parameters from data containing severe outliers, a common challenge in computer vision.

Research Potential 

The proposed LSC method demonstrates substantial advantages in handling complex datasets, improving model fitting accuracy and speed, and providing technical support and solutions for multiple fields such as computer vision, autonomous driving, robot navigation, AR/VR, and motion analysis.

Methodology

ž The principle of LSC is to retain the latent semantic consensus between data points and model hypotheses.

ž Specifically, LSC constructs separate latent semantic spaces for data points and model hypotheses and transforms the model fitting problem into these two latent semantic spaces.

ž Then, LSC explores the distribution of points in these latent semantic spaces to remove outliers, generate high-quality model hypotheses, and effectively estimate model instances.

ž Finally, due to its deterministic fitting characteristics and efficiency, LSC can provide consistent and reliable solutions for general multi-structure model fitting within a few milliseconds.

The team tested the performance on both synthetic data and real images. Compared with several state-of-the-art model fitting methods, the proposed LSC method achieved significant improvements in both accuracy and speed.

Original Thesis

Latent Semantic Consensus for Deterministic Geometric Model Fitting | IEEE Journals & Magazine | IEEE Xplore


We use cookies to offer you a better browsing experience, analyze site traffic and personalize content. By using this site, you agree to our use of cookies. Visit our cookie policy to leamn more.