Non-parametric active target localization: Exploiting unimodality and separability
Dhruva Mokhasunavisu, Urbashi Mitra · 2017
The problem of localizing the peak of a unimodal signal from noisy measurements is examined. A non-Bayesian framework is used to identify feasible peak locations based on collected measurements in the absence of prior information regarding the signal. An active sensing algorithm is designed to reduce the number of measurements without compromising the localization accuracy. The sensing algorithm exploits the unimodality of the signal using a group testing strategy to iteratively reduce the ambiguity associated with the location of the target while adaptively excluding sample locations that are irrelevant for localization. While it can be numerically demonstrated that the greedy approach is not always optimal; we can bound this probability and it is found to be small. Analysis and experiments suggest that the greedy step is optimal with high probability in most cases. Furthermore, it can be shown that using a greedy approach achieves better localization performance than using all the available samples.