Discrimination gain to optimize detection and classification

Abstract

A method for managing agile sensors to optimize detection and classication based on discrimination gain is presented. Expected discrimination gain is used to determine threshold settings and search order for a collection of discrete detection cells. This is applied in a low signal-tonoise environment where target-containing cells must be sampled many times before a target can be detected or classied with high condence. The goal of sensor management i s i n terpreted to be to direct sensors to optimize the probability densities produced by a data fusion system that they feed. The use of discrimination is motivated by its interpretation as a measure of the relative likelihood for alternative probability densities. This is studied in a problem where a single sensor can be directed at any detection cell in the surveillance volume for each sample. Bayes rule is used to construct a recursive estimator for the cell target probabilities. The expected discrimination gain is predicted for each cell using its current target probability estimates. This gain is used to select the optimal cell for the next sample. For thresholded data, the expected discrimination gain depends on the threshold which is selected to maximize the gain for each sample. The expected discrimination gains can be maintained in a binary search tree structure for computational eciency. The computational complexity of this algorithm is proportional to the height of the tree which is logarithmic in the number of detection cells. In a test case for a single 0 dB Gaussian target, the error rate for discrimination directed search w as similar to the direct search result against a 6 dB target.

Towards Contextual Action Recognition and Target Localization with Active Allocation of Attention

“…The main assumption is that the observer can use features which allow to discriminate between the different affordances and the effector. This approach is similar to that of [8] and [18] but, instead of being limited to track or to find objects in a dynamic environment, it allows for the active recognition of a dynamic event.…”

An Information Roadmap Method for Robotic Sensor Path Planning

“…Several informationtheoretic functions have been proposed for this purpose. Cross entropy was used in [28] to solve a multisensor-multitarget assignment problem, and in [29,30] to manage agile sensors with Gaussian models for target detection and classification. Entropy and the Mahalanobis distance measure were used in [27] for sensor selection in ad-hoc sensor networks.…”

“…While robot path planning typically aims to optimize a deterministic additive function such as Eucledian distance, sensor planning aims to optimize a stochastic sensing objective that is not necessarily additive. Another basic difficulty in sensor planning is that, although the measurements ultimately determine the sensor performance, they cannot be factored into the planning problem because the sensor's position must be planned prior to obtaining the sensor's measurements [27][28][29][30][31]. Recently, several authors have shown that this difficulty can be overcome by an approach known as information-driven sensor planning, which uses information theoretic functions to estimate the measurements' value prior to deploying the sensor [27][28][29][30].…”

Distributed sensor resource management and planning

“…Most of the literature is in the area of managing sensors to maximize kinematic information gain only [1][2]. Some prior art exists in managing sensors for maximizing ID and search as well [3][4]. One can use information theoretic criteria such as entropy, discrimination information, mutual information, etc.…”

“…Hintz and McVey [45] used entropy for search, track and ID tasks. Work in [46][47][48][49][50][51][52][53] use information measures such as entropy and discrimination gain for goals such as determining resolution level of a sensor, determining priority of search and track tasks, etc. Hintz et al [9,10] use the Shannon entropy, as we do in this paper, while Schmaedeke and Kastella [11] have chosen to use Kullback-Leibler (KL) divergence as measure of information gain.…”