Conformal Prediction for Trustworthy Detection of Railway Signals

TL;DR

This paper applies conformal prediction to railway signal detection, providing uncertainty guarantees and improving bounding box reliability in a novel train operator perspective dataset.

Contribution

It introduces the use of conformal prediction for bounding box correction in railway signal detection, enhancing trustworthiness of ML models in this domain.

Findings

Conformalization improves bounding box accuracy with probabilistic guarantees.

The method is validated on a novel dataset from train operator perspective.

Results show increased reliability of railway signal detection models.

Abstract

We present an application of conformal prediction, a form of uncertainty quantification with guarantees, to the detection of railway signals. State-of-the-art architectures are tested and the most promising one undergoes the process of conformalization, where a correction is applied to the predicted bounding boxes (i.e. to their height and width) such that they comply with a predefined probability of success. We work with a novel exploratory dataset of images taken from the perspective of a train operator, as a first step to build and validate future trustworthy machine learning models for the detection of railway signals.