Predictive multiple sampling algorithm with overlapping integration intervals for linear wide dynamic range integrating image sensors

摘要

Machine vision systems are used in a wide range of applications such as security, automated quality control and intelligent transportation systems. Several of these systems need to extract information from natural scenes in the section of the electromagnetic spectrum visible to humans. These scenes can easily have intra-frame illumination ratios in excess of 10⁶ : 1. Solid-state image sensors that can correctly process wide illumination dynamic range scenes are therefore required to ensure correct reliability and performance. This thesis describes a new algorithm to linearly increase the illumination dynamic range of integrating-type image sensors. A user-defined integration time is taken as a reference to create a potentially large set of integration intervals of different duration (the selected integration time being the longest) but with a common end. The light intensity received by each pixel in the sensing array is used to choose the optimal integration interval from the set, while a pixel saturation predictive decision is used to overlap the integration intervals within the given integration time such that only one frame using the optimal integration interval for each pixel is produced. The total integration time is never exceeded. Benefits from this approach are motion minimization, real-time operation, reduced memory requirements, programmable light intensity dynamic range increase and access to incremental light intensity information during the integration time.