Object Detection Methods for Optical Survey Measurements Alejandro Pastor GMV, apastor@gmv.com Diego Escobar GMV, descobar@gmv.com Manuel Sanjurjo-Rivo Universidad Carlos III de Madrid, manuel.sanjurjo@uc3m.es Alberto ´ Agueda GMV, aagueda@gmv.com ABSTRACT This work presents a novel sequential filtering algorithm able to identify new objects from optical measurement data by associating uncorrelated tracks (UCTs) belonging to the same object. It makes use of both Initial Orbit Determination (IOD) and Orbit Determination (OD) methods to evaluate a figure of merit to help deciding whether certain tracks belong to the same object or not. Instead of using a brute-force approach by evaluating all possible combinations of UCTs, several filters and complexity reduction techniques are used to reduce the computational resources required. Furthermore, the association is performed on the measurements space (track-to-track correlation) rather than in the orbit space (track-to-orbit or orbit-to-orbit correlation). A generic object detection methodology based on track association is presented, with a special emphasis on the correlation of optical tracks, although it has been already applied to correlation of radar tracks. In the former case, the problem resides in the derivation of enough orbital information from a single track (with less attributables than a radar track) to allow the application of filters and complexity reduction techniques. Hence, optical track association is far more complex and resource-consuming than the radar track association, and thus classical approaches are not enough. Results have shown that this strategy provides more reliable results than an association made on the orbit space, in terms of both false positives and number of missed objects. A realistic simulated scenario has been set up to evaluate the performance of the correlation procedure under a purely build-up scenario. The performance of the methods is evaluated in terms of clear and well-defined correlation metrics, such as true positives, false positives and false negatives, providing both ratios and absolute values. They prove that the proposed methodology is able to provide excellent results for the track association problem, since most of the objects can be detected while providing a very low number of false detections. This is important during catalog build-up, since the addition of wrong objects is very undesirable. The computational cost of the algorithm allows real-time processing of new tracks thanks to the selective generation and pruning that avoid evaluating all possible combinations of tracks. 1. INTRODUCTION The number of resident space objects (RSO) is increasing year after year and therefore the sensing capabilities are also growing [1]. Space Surveillance and Tracking (SST) systems are composed by sensors and on-ground processing infrastructure devoted to the generation of a catalog of RSO: a robust automated database that contains information of every detected object. During surveillance, large areas of the sky are scanned to obtain data for both catalog build-up and maintenance activities. The catalog build-up process consists in detecting new objects to include them in the catalog without any previous information, while the maintenance task entails the update of existing objects information. Hence, the catalog build-up depends on the capability to detect new objects from measurements, packed as tracks, provided by a sensor network. Copyright © 2019 Advanced Maui Optical and Space Surveillance Technologies Conference (AMOS) – www.amostech.com