A Fusion Analysis and Evaluation Tool for Multi-Sensor Classification Systems

Fusion of information from multiple sources to achieve performances exceeding those of individual sources has been recognized in diverse areas [17] such as reliability, forecasting, pattern recognition, neural networks, decision fusion, and statistical estimation. In engineering systems, the fusion methods have proven to be particularly important since they can provide system capabilities with multiple sensors significantly beyond those of single sensor systems. Multi-sensor data fusion allows the combination of information from sensors with different physical characteristics to enhance the understanding of the surroundings and provide the basis for planning, decision-making, and control of autonomous and intelligent machines. It seeks to combine information from multiple sensors and sources to achieve inferences that are not feasible from a single sensor or source. To fully exploit the capabilities of a fusion system, modeling and performance evaluation methodologies are critical in order to optimally design and effectively evaluate fusion performance of multiple heterogeneous sensor data. In particular, a systematic approach to evaluate the overall performance of the system is indispensable. To allow developers and users to assess their fusion system performance under various conditions before a data fusion system is deployed, a tool based on the Fusion Performance Model (FPM) [8] was developed with a focus on one of the most important performance measures, spatial and classification performance modeling and prediction. Note that the purpose of the FPM is to predict performance given sensor suite and operating conditions. For a sensor fusion system, typical questions that could be asked would be “what is the best achievable performance, and is it good enough?” The FPM will be able to answer the first question and if the answer is “not good enough,” a sequence of “what if” scenarios can be added for FPM to conduct new assessments. Those scenarios may include changing operating conditions, such as signal-to-noise ratio (SNR), geometry, and revisit rate, to name a few of the existing sensors or adding new sensors. The assessment results can then be used to better manage sensors and allocate system resources. While the FPM model described in [8] developed a kinematic performance prediction methodology and defined the classification performance model and [7] described an analytical method to predict classification performance and an efficient approximate formula to estimate the average probability of correct classification given sensor characteristics, there is still a lack of effective tools to evaluate a fusion system performance as described in [8] and [7] in an easy and accessible way in order to make the assessment results promptly available to better control sensors and allocate system resources.