Segmentation and linguistic summarization of voxel environments using stereo vision and genetic algorithms

For reasons such as computational complexity, spatial and temporal information reduction, and human understandability, it is important that computer vision systems be equipped with the means to summarize their content in a natural language. Such rich descriptions are of use by both humans and computers for describing, recognizing, and tracking objects, activity, and their interactions at a desired level of abstraction. A genetic algorithm is introduced here for segmenting non-human objects deemed relevant to human activity analysis in stereo vision acquired voxel environments. This approach is of use in an Eldercare setting as it relates to monitoring the “well-being” of residents through acquiring and detecting deviations in patterns of typical behavior as well as recognizing abnormal events, such as fall detection.