Strategies for Fusion at the Detection Level Sample Clauses

Strategies for Fusion at the Detection Level. Detection-to-Detection Fusion Fusing detections to detections has got the advantage that the resulting composite detections can be fed into any exist- ing (unimodal) tracking algorithm, without requiring special provisions to cope with information from multiple detectors. This approach can also be helpful if the tracking algorithm itself is computationally very complex and scales badly with the number of detections. For detection-to-detection fusion, we have im- plemented a series of nodelets6 which can be used to flexibly compose a fusion pipeline by means of roslaunch XML files: – Converter nodelets for converting DetectedPersons messages into Compos- iteDetectedPersons messages, and vice versa. – Aggregator nodelets which simply concatenate two sets of CompositeDetect- edPersons. Useful for combining detections from sensors with non-overlapping fields of view, in which case no data association is required. – Fusion nodelets that perform data association between two sets of Composite- DetectedPersons, e. g. using a nearest-neighbor data association technique. 6 A powerful mechanism to dynamically combine multiple algorithms into a single process with zero-copy communication cost. See xxxx://xxxx.xxx.xxx/nodelet. 16 People Detection, Tracking and Visualization on a Mobile Robot