Transcript Document
Multi-Sensor Data Fusion H.B. Mitchell UNCLASSIFIED Multi-Sensor Data Fusion Three-hour tutorial on multi-sensor data fusion The tutorial is closely based on a selection of material taken from the book: Multi-Sensor Data Fusion: An Introduction by H.B. Mitchell published by Springer-Verlag (2007) 1. Introduction 2. Sensors 3. Common Representational Format 4. Spatial, Temporal, Semantic Alignment 5. Robust Statistics 6. Ensemble Learning UNCLASSIFIED Introduction UNCLASSIFIED Multi-Sensor Data Fusion Data fusion: Theory and Techniques which combine sensor data into a common representational format. Aim is to improve the quality of information. Data fusion is analogous to the manner in which humans and animals improve their chances of survival by exploiting their Multiple senses Experience Ability to reason Man-Made Fusion System EO, IR, Radar A priori information and/or historical information Bayesian inference, fuzzy logic, Dempster-Shaefer UNCLASSIFIED Multi-Disciplinary Subject Multi-sensor data fusion brings together many different techniques and applications Techniques Bayesian networks Signal Processing Statistical Estimation Tracking Algorithms Classification Algorithms Invariant Subspaces MCMC Genetic Algorithms Bagging, Boosting Computing Power Applications Fusion UNCLASSIFIED Medical Imaging Remote Sensing Surveillance Data Mining Computer Vision Stereo Imaging Complementary Fusion After. Toet. Natural color mapping for multiand nightvision imagery. Information Fusion (2003) UNCLASSIFIED Pan Sharpening Panochromatic image Multi-spectral image UNCLASSIFIED Pan-sharpened image Colorization After. Toet. Natural color mapping for multiband nightvision iamgery. Information Fusion (2003) UNCLASSIFIED Sensors UNCLASSIFIED Sensors Sensors are devices which interact directly with environment Sensors are the source of all input data. Often use smart sensors which • • • Transform sensor signal to standardized digital format Calibrates sensor signal Transmits sensor signal via standardized interfaces. sensor element Amp UNCLASSIFIED Filter transmitter/ receiver A/D m Fusion Node Build a Data Fusion System as a distributed assembly of fusion nodes Input Data Fusion Output Data Aux Inform. Ext. Inform UNCLASSIFIED Logical Sensor Logical sensor is any device which functions as a source of inform. for a multi-sensor data fusion node S1 S2 F1 F2 S1: Physical sensor F1: Virtual Sensor: A fusion node whose output is fed into another fusion node “virtual” sensor S3 UNCLASSIFIED Sensor Errors Sensors only give an estimate of the measured physical property Nature of errors often determine the preferred fusion algorithm Bias. Separately track with each set of measurements, then fuse tracks. No Bias. Concatenate measurements into one vector then track with Kalman Filter UNCLASSIFIED Common Representational Format UNCLASSIFIED