Selective sensor control for multi-object tracking

Sensor control for multi-object tracking is a challenging problem, as for a given observation set, the detected measurements are interlaced with false alarms and there is no certainty about the object that generated the measurement. This is a widely discussed topic in the recent research literature and is usually treated as an optimisation problem that aims to obtain the best measurements for multi-object tracking purposes, by way of selecting the best control command from a set of employable commands. The novelty of this research work lays in enabling selectivity in sensor control which imparts the ability to selectively track specific targets in a group using the recent Label Multi-Bernoulli filters, and this perspective has not been discussed in any previous literature. The goodness of sensor measurements is a defining factor in such applications, and the aim of sensor control is to lead towards better measurements for tracking.

Recent stochastic filtering methods based on Random Finite Sets are utilised for this work by the use of Labeled Multi-Bernoulli filters. New sensor control methods have been formulated for selective tracking in a multi-target single sensor scenario and have been extended for multi-target multi-sensor applications. The significant research gaps in the field were identified and this work aims at bridging those gaps, and is aimed at devising selective accelerated sensor control methods for tracking applications where only a selected subset of targets in the surveillance area is of primary interest.

This document is organised as follows: the first chapter briefly elaborates the aims and objectives of the research and the motivation behind it. The background information on target tracking and sensor control is provided in the second chapter. Further, the foundations for sensor control and selective multi-target tracking are built through a brief literature review in the third chapter. This is followed by a discussion on the implementation of selective sensor control with Labeled Multi-Bernoulli filter for a single sensor, in the fourth chapter. The methods developed for selective multi-sensor control, accelerated solutions, simulations and results are discussed in the fifth chapter. The sixth chapter discusses the performance analysis of the proposed accelerated sensor control solution with real-time on-line sensor data. The seventh chapter closes the thesis by presenting the summary of contributions, conclusions and future research directions. The bibliography is provided at the end.

This thesis is in agreement with the RMIT University rules for the content and format of a thesis and is presented as a series of contributions in the form of publications delivered during the course of this research work.