Multisensor Data Fusion (SIP 7005)

Multisensor Data Fusion
Subject code: SIP 7005

Lecturers
Dr Branko Ristic and Dr Neil Gordon
DSTO

Mode of delivery
On-line with a possibility of weekly lectures.

Assumed knowledge
Linear algebra (matrices), probability theory, estimation theory, Kalman Filtering and Tracking (SIP 7002), MATLAB.

Aim/Learning Objectives
To provide the student with the theoretical and practical knowledge of an emerging field of data fusion, mainly in the context of modern surveillance systems. The emphasis will be on target tracking and identification in a network centric environment, covering techniques used for filtering, measurement association, track association and fusion, sensor registration and fusion based target identification.

Content
Overview: The role of multiple sensor fusion; Typical applications and sensors; Benefits of information fusion; Problems and limitations.

Architectural concepts and network issues: Centralised, distributed and hybrid architectures; Typical network issues (communication bandwidth, data latency, data incest, picture consistency)

Centralised multi-sensor filtering: Alternative track update methods; Filtering out-of-sequence measurements (algorithm A and B); Performance bounds for filtering.

Distributed multi-sensor filtering via track fusion: Bar-Shalom-Campo fusion; Information matrix fusion; Fusion using equivalent measurements.

Data association methods (measurement-to-track association in centarlised architectures and track-to-track association in distributed architectures): Distance measures; Gating; Global nearest neighbour; Joint probabilistic data association.

Sensor Registration: Registration error sources; Least squares method; Recursive methods; Coordinate systems.

Fusion of target ID declarations: Problem description; Heuristic methods; Bayesian inference; Evidential theory approach; A case study.

Assessment
Assignment (60%) and examination (40%), however the percentages are indicative only. Details can be found in the study guide provided at the start of the semester.

Assignment projects:

The topic of assignment project should be one of the following:

Track fusion and comparison with the centralised architecture.
Sensor registration.
Data association.
Filtering out-of-sequence measurements.
Distributed multiple passive sensor tracking.
Fusion of active (position) and passive (angle-only) tracks.
Fusion of target ID declarations.

Students can also make their own suggestion of a topic. The assignment topic has to be defined and approved by the lecturer before week 4. The assignment reports must include a literature review, a description of the problem and its algorithmic solution(s), implementation, numerical simulations with performance evaluation. Some comparisons and conclusions are highly desirable.

Resources
All the materials necessary for the course will be availabe on-line. The lecture notes also include an extended bibliography for further reading on the subject.