TRAVELING SCENE RECOGNITION MODEL GENERATION DEVICE, DRIVING SUPPORT DEVICE, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To compress a traveling scene while retaining pieces of information required for driving even when an object to be recognized is not specified in advance.

SOLUTION: A sight-line end image extraction part 34 extracts a sight-line end image on the basis of the sight-line direction of a driver, an extraction part 36 of feature quantity A generates a feature quantity list A from the sight-line end image, and an extraction part 38 of feature quantity B generates a feature quantity list B from an entire exterior environment image and accumulates it in a feature quantity DB 40. A cluster DB construction part 42 performs clustering of the feature quantity list A and stores a representative vector of each cluster in a cluster DB 44. A cluster weight DB construction part 46 calculates a weight for each cluster on the basis of a ratio of appearance probability of the feature quantity list A to appearance probability of the feature quantity list B in each cluster, and stores it in a cluster weight DB 48. A scene recognition part 52 calculates an appearance frequency histogram for each cluster of a feature quantity list C to clusters having a cluster weight equal to or more than a threshold, and recognizes it as a traveling scene.

COPYRIGHT: (C)2013,JPO&INPIT