1A1-I01 Proposal of the stationary obstacle detection method by combining estimation of object position using the clustering method with image recognition using machine learning

Abstract

We propose the method to detect a stationary obstacle in the environment by a laser range finder and a monocular camera.The method is divided into two systems, the estimation of object system and the image recognition system.The system for estimation of object position uses k- means clustering, and improves disadvantage of k-means clustering that does not guarantee to split clusters optimally depending on the situation.The system for image recognition uses the method of Bag of Features to represent image features and the method of Support Vector Machine in classification.Both systems are run multi-threaded, maintain the real-time properties.Then, we evaluate the performance of each system and confirm behavior when both systems are combined.