Error models for light sensors by statistical analysis of raw sensor measurements

Error modeling is a procedure of quantitatively characterizing the likelihood that a particular value of error is associated with a particular measured value. Error modeling directly affects accuracy and effectiveness of many tasks in sensor-based systems including calibration, sensor fusion and power management. We developed a system of statistical techniques that calculate the likelihood that error of a particular value is part of a measurement. The error modeling approach has three steps: (i) data set partitioning; (ii) constructing the error density model; and (iii) learn-and-test and resubstitution-based procedures for validating the models. The data set partitioning identifies a specified percentage of measurements that have the highest negative discrepancy between sensor and standard measurements. The partitioning step employs data fitting models to identify compact curves that represent the partitioned subsets. The error density modeling uses the compact curves to build the probability density function (PDF) of the error. For validation purposes, we use a resubstitution-based paradigm.