An Occupant Classification System Eigen Shapes or Knowledge-Based Features

摘要: This paper describes an occupant classification system based on Eigen shapes and support vector machines using 3D data. The is used to classify type into adult, a child, child seat, or object. inputs are depth images from time-of-flight camera. first segmented the background normalized for threedimensional translations. projected that constructed training set. projections as feature vectors algorithm. features complement each other since former linear transformation reduce dimensionality of space, while latter can deduce non-linear aspects these lower-dimensional features. Our experiments lead several conclusions. First, we compare between knowledge features, e.g. computer generated versus human provides better results compared various combinations knowledge-based combination provide best results, both capture different characteristics images. Comparison with intensity analysis shows more suitable this application. have also been shown be superior algorithms. than 98 percent recognition rate input. failure cases include extreme deformations were not part