Taizo Umezaki's research while affiliated with Chubu University and other places

This paper proposes methods for unsupervised lexical acquisition for relative spatial concepts using spoken user utterances. A robot with a flexible spoken dialog system must be able to acquire linguistic representation and its meaning specific to an environment through interactions with humans as children do. Specifically, relative spatial concepts (e.g. front and right) are widely used in our daily lives, however, it is not obvious which object is a reference object when a robot learns relative spatial concepts. Therefore, we propose methods by which a robot without prior knowledge of words can learn relative spatial concepts. The methods are formulated using a probabilistic model to estimate the proper reference objects and distributions representing concepts simultaneously. The experimental results show that relative spatial concepts and a phoneme sequence representing each concept can be learned under the condition that the robot does not know which located object is the reference object. Additionally, we show that two processes in the proposed method improve the estimation accuracy of the concepts: generating candidate word sequences by class n-gram and selecting word sequences using location information. Furthermore, we show that clues to reference objects improve accuracy even though the number of candidate reference objects increases.

This paper proposes methods for unsupervised lexical acquisition for relative spatial concepts using spoken user utterances. A robot with a flexible spoken dialog system must be able to acquire linguistic representation and its meaning specific to an environment through interactions with humans as children do. Specifically, relative spatial concepts (e.g., front and right) are widely used in our daily lives, however, it is not obvious which object is a reference object when a robot learns relative spatial concepts. Therefore, we propose methods by which a robot without prior knowledge of words can learn relative spatial concepts. The methods are formulated using a probabilistic model to estimate the proper reference objects and distributions representing concepts simultaneously. The experimental results show that relative spatial concepts and a phoneme sequence representing each concept can be learned under the condition that the robot does not know which located object is the reference object. Additionally, we show that two processes in the proposed method improve the estimation accuracy of the concepts: generating candidate word sequences by class n-gram and selecting word sequences using location information. Furthermore, we show that clues to reference objects improve accuracy even though the number of candidate reference objects increases.

Nurse’s work is the task that leads to a serious accident with one mistake or miss, and the nurses are exposed to high stress. In particular, injection / pre-medication, tube connection, and falling are factors that lead to serious accidents and are considered to be a major load factor of nursing work. In order to reduce the burden of nursing work, we are working on the development of a sensing system to prevent fall accidents. Since fall accidents tend to occur when elderly people whose lower limb muscle strength has declined go to the toilet, we use the camera image to detect the end position, which is the initial posture of the patient’s landing movement. In this paper, we detected the sitting position of the patient by combining the detection result of the skeletal position of the patient and the detection result of the bed position. A simulation environment was constructed and the estimation accuracy of the end sitting position of the patient was evaluated by using the image taken of the scene where the patient and the nurse are active.

The work of a nurse involves tasks that can lead to serious accidents with a single mistake or miss, and thus nurses are exposed to high stress. In particular, injections, pre-medications, tube connections, and falling are factors that lead to serious accidents and are considered a major part of the load of nursing work. To reduce the burden of nursing work, we are working on developing a sensing system to prevent fall accidents. Because fall accidents tend to occur when elderly people, whose lower limb muscle strength has declined, go to the toilet, we use a camera image to detect the end position, which is the initial posture of the patient's landing movement. In this study, we detected the sitting position of the patient by combining the detection result of the skeletal position of the patient and the detection result of the bed position. A simulation environment was constructed and the estimation accuracy of the end sitting position of the patient was evaluated using the image captured at the scene where the patient and the nurse are active.

We propose a brand-new measurement method for glossy surfaces by the unique lighting module with a transmission type holographic optical element (HOE). Since the diffracted lights from HOE has large chromatic dispersion, the diffraction direction is determined by colors. This characteristic means that light ray directions from glossy surfaces can be analyzed using chromatic property in captured image. In our method, hue value is treated as a function of light ray direction. Therefore, normal direction on a surface in each pixel can be estimated by using hue value in the same pixel. In this paper, relations between hue values and light ray directions are estimated using approximated formula and we discuss the calculation result for sample objects.

In hospitals, in order to prevent a patient from falling and fall accidents, there is an increasing demand to detect a lifting movement and a leaving‐off movement of a patient with a camera installed in a hospital room. As a measure for efficiently detecting these operations, a process of specifying a bed position which is an occurrence position of an action is effective. We propose a method to detect bed position from image of monocular camera, using image features and framework of machine learning. By generating a viewpoint conversion image for an input image, a uniform bed shape is acquired even for images at different camera positions. Evaluation was made with 2160 still images and 3888 scene moving images to confirm the usefulness of the method.

In recent years, as a way to prevent patient fall-down, studies have been conducted using patient room cameras to detect the patient behavior of leaving the bed. It is very important to specify the patient bed location in the process of detecting patient behavior using camera images. In this study, we propose a method to specify the patient bed location using a monocular camera. In this proposal, we convert a camera image viewpoint into a bird's-eye view image as a preprocessing step. By using planer perspective transformation, it is possible to display the bed as a rectangular shape with a fixed ratio, even if the bed location or camera position is changed. Therefore, it is possible to detect the bed location with a high degree of accuracy by means of machine learning. The simulation experiment results confirm that the average error and standard deviation of the bed coordinates are 7.9 and 5.0 pixels, respectively; in the practical scene, we confirm that the average error and standard deviation of the bed coordinates are 12.1 and 8.2 pixels, respectively.

In hospitals, in order to prevent a patient from falling and fall accidents, there is an increasing demand to detect a lifting movement and a leaving-off movement of a patient with a camera installed in a hospital room. As a measure for efficiently detecting these operations, a process of specifying a bed position which is an occurrence position of an action is effective. We propose a method to detect bed position from image of monocular camera, using image features and framework of machine learning. By generating a viewpoint conversion image for an input image, a uniform bed shape is acquired even for images at different camera positions. Evaluation was made with 2160 still images and 3888 scene moving images to confirm the usefulness of the method.

A lighting module using a holographic optical element (HOE) for glossy surface inspection has been developed and demonstrated. This lighting module has several advantages for such kind of inspection compare to the usual lighting system. The differences between HOEs lighting system and usual lighting system were discussed. HOEs were fabricated by our originally developed ultra-high speed collinear holographic printer. The several optical systems and image processing method using hue also were developed, the validity of our system for actual defect was investigated.

Nowadays, the research that robots learn the meaning of a word by relating speech instructions with signals of the sensorimotor system has started. Lots of pre-research on learning ‘concept’ that means the meaning of a word by robot have used the co-occurrence information of words and sensor signals directly. However, relative concepts, such as color, size or location, require comparing with other objects or concepts. The objects or concepts used for comparison are called “reference object.” In human-computer interaction, reference point may not be explicitly communicated, because human utterance is ambiguous. Therefore, estimating a true reference object from the multi-candidate is necessary to learn relative concepts. In this paper, we propose a learning method based on EM algorithm to estimate the reference point and learn relative concepts. In our experiments, a robot learned concepts expressing relative location, such as ‘in front of or ‘behind.’ The robot was given text labels expressing its locations from a human teacher in several situations. Nevertheless, the instructions didn’t include the true reference points and the coordinate system of each concept. Under these conditions, the robot could learn the concepts based on coordinate systems have the orientations of the reference objects or the teacher.