Hirofumi Kanazaki

Supervised Learning for Object Classification from Image and RFID Data

Position estimation and tracking of multiple objects by vision sensors is one of the most fundamental technologies. While the vision sensors provide high accuracy measurements for position estimation, they require suitable features of objects for accurate recognition and detection as prior knowledge. Especially, learning of appearance based features of objects requires large quantities of training data, which makes development costs. This paper proposes a method for learning appearance based features of objects using auxiliary data of RFID. In this method, the RFID device is used as a supervisor to semi-automatically construct the training data set for each object. Since it is difficult to observe what ID does an object image correspond to, this problem comes down to supervised learning using incompletely labeled features. This paper proposes a learning method using Kernel PCA and EM algorithm, and verifies the effectiveness and robustness of this method

Bearing-only mapping by sequential triangulation and multi-dimensional scaling

In this paper, we introduce an alternative solution to the bearing-only mapping problem in which a mobile robot builds a map of features (landmarks) using only relative bearing measurements to them and odometry information. Our approach named BOM-STMDS (bearing-only mapping by sequential triangulation and multi-dimensional scaling) first tries to estimate relative distances among the features, then finds two-dimensional coordinates of all features by using multi-dimensional scaling (MDS) and its enhancements. BOM- STMDS is different from the conventional BOSLAM methods based on Bayesian filtering in that robot self-localization is not mandatory. Another remarkable property is that BOM-STMDS is able to utilize prior information about relative distances among features efficiently. In the experiment, the performance of BOM-STMDS is shown to be competitive with a conventional EKF-based BOSLAM method.

Map building without localization by estimation of inter-feature distances

This paper proposes an alternative solution to a mapping problem in two different cases; when bearing measurements to features (landmarks) and odometry are measured and when bearing and range measurements to features are measured. Our approach named M-SEIFD (Mapping by Sequential Estimation of Inter-Feature Distances) first estimates inter-feature distances, then finds global position of all the features by enhanced multi-dimensional scaling (MDS). M-SEIFD is different from the conventional SLAM methods based on Bayesian filtering in that robot self-localization is not compulsory and that M-SEIFD is able to utilize prior information about relative distances among features directly. We show that M-SEIFD is able to achieve a decent map of features both in simulation and in real-world environment with a mobile robot.

Map Building by Sequential Estimation of Inter-feature Distances

This paper proposes an alternative solution to a mapping problem in two different cases; when bearing measurement to features (landmarks) and odometry are measured and when local position of features are measured. Our approach named M-SEIFD (Mapping by Sequential Estimation of Inter-Feature Distances) first estimates inter-feature distances, then finds global position of all features by enhanced multi-dimensional scaling (MDS). M-SEIFD is different from the conventional SLAM methods based on Bayesian filtering in that robot self-localization is not compulsory and that M-SEIFD is able to utilize prior information about relative distances among features directly. We show that M-SEIFD is able to achieve a decent map of features both in simulation and in real-world environment with a mobile robot.

Probabilistic Sensor Models for Multiple Objects Localization Problem

This research intends to build up the probabilistic models of three heterogeneous sensors, camera, range sensor and RFID for dealing with objects localization problems, and also aims to fulfil the sensor fusion with these models. In this paper, we describe the mathematical significance for the sensor models, and define concrete models. Then we show the effectiveness of sensor fusion with these models from the simulation