TRAVERSABILITY TERRAIN IMAGE RECOGNITION WITH MULTI-WEIGHT VECTOR PROJECTION METHOD

摘要

This paper develops a supervised method for traversability terrain image recognition.In the traversability terrain Image recognition task, a good recognition algorithm should have not only high recognition performance but also high efficiency. This proposed method is developed for this purpose, which is based on our recent work.This central idea is to seek to find multiple weight-vector projection directions for each class, such that the projected points of one class can be well separated from those of the other class.The proposed method is evaluated on two outdoor scene image databases.The experimental results disclose that our proposed method outperforms other classification algorithms.