Spatio-Temporal Alignment between Cooperative Sensor Platforms

Abstract

Cooperative perception, while increasingly feasible, remains a complex topic. One of its key challenges is the spatio-temporal alignment of loosely coupled sensor platforms. This is particularly difficult when independent parties maintain the platforms and only exchange information via V2X messages. In this context, we assume spatial misalignment and a signifi- cant deviation between the platform's unsynchronized clocks. We propose a novel method to address the problem of spatio-temporal alignment. Our method is unique in that it solves this issue without prior knowledge but the detected objects provided by both sensor platforms. Our solution applies a Gaussian Mixture Model on the locally detected objects to maximize the likelihood of the objects retrieved via V2X communication. We furthermore interpolate this probability density function using tracking information of the locally detected objects, which enables us to optimize for the spatial transformation and estimate the clock's deviation. We applied our method to synthetic and real-world data recorded at an intersection in Ingolstadt, Germany, significantly reducing the average Euclidean distance between matching objects. This practical application has not only validated our approach but also opened up possibilities for further cooperative perception tasks.