Managing Dynamism of Multimodal Detection in Machine Vision Using Selection of Phenotypes

Multimodal Sensor Vision is a technique for detecting objects in dynamic and uncertain environmental conditions. In this research, a new approach for automated feature subset selection-mechanism is proposed that combines a set of features acquired from multiple sensors. Based on changing environmental conditions, the merits of respective sensory data can be assessed and the feature subset optimized, using genetic operators. Genetic Algorithms (GAs) with problem specific modifications improve reliability and adaptability of the detection process. In the new approach, a traditional GA is customized by combining the problem profiled encoding with a specialized operator. Application of an additional operator prioritizes and switches within the feature subsets of the algorithm, allowing a feature level aggregation that uses the most prominent features. The approach offers a more robust and a better performing Machine Vision processing