From Measure Theory to Machine Learning: Product Coefficients for LiDAR Data

In this talk, I present our ongoing work on improving 3D LiDAR point-cloud classification using product coefficients, mathematical quantities derived from measure theory that describe the local structure of point distributions. In our earlier study, we introduced these coefficients and showed how adding them as features alongside PCA improved classification compared to standard approaches. Our recent work extends this idea by combining product coefficients with autoencoders and a K-Nearest Neighbors classifier, and by exploring how adding coefficients level by level influences performance.