This paper proposes a tracking method of objects, such as cars, two-wheelers, and pedestrians, using multiple light detection and ranging sensors (LiDARs) set in an intersection environment. Each LiDAR unit detects objects using deep learning CenterPoint from its own LiDAR point cloud data and exchanges object information (i.e., the position, size, heading angle, and confidence score of the objects) by communicating with its neighboring LiDAR units. Then, the object information is fused. Here, each LiDAR unit estimates the object poses, and the estimates are fused by exchanging the information among the neighboring LiDAR units. A distributed interacting multimodel estimator is employed to accurately estimate the poses of objects under various motion modes, such as stopping and suddenly moving and stopping, in a distributed manner without requiring a central server. Simulation experiments using four LiDAR units with a mesh type of network topology set at signal poles in an intersection environment validate the performance of the proposed method.

This paper presents a cooperative method for tracking objects, such as cars, two-wheelers, and pedestrians, using multiple light detection and ranging sensors (LiDARs) deployed in a road environment. Each LiDAR system independently estimates the motion and shape of objects, e.g., in terms of their positions and velocities, and sizes, within its field of view. A central server collects and fuses these estimates from LiDAR systems. The proposed approach uses an extended object-tracking method based on the multiplicative error modelextended Kalman filter (MEM-EKF) framework combined with an interacting multiple model estimator to accurately track the motions and shapes of objects exhibiting various movement patterns, such as stopping, constant velocity motion, and suddenly moving and stopping. In addition, information filterbased fusion is performed at the central server. Simulation results using two LiDAR systems in an intersection environment demonstrate the effectiveness of the proposed method compared to the conventional single-model MEM-EKF method and standalone tracking method with a single LiDAR system.

This paper presents an object recognition and tracking method using multiple light detection and ranging sensors (LiDARs) set in an environment. Each LiDAR recognizes objects, such as cars, two-wheelers, and people, using CenterPoint, which is an end-to-end method for three-dimensional object recognition. Information about recognized objects (the position, size, heading, and the confidence score of the objects) is sent to a central server and fused using nonmaximum weighted method. The fused objects are tracked using an interacting multimodel estimator, and their poses and behavior are estimated. Simulation experiments using three LiDARs placed in an intersection environment validate the performance of the presented method.

This paper presents a cooperative people tracking using two networked lidars allocated in an environment as sensor nodes. The tilt angles, heights, and relative poses of the two lidars are calibrated based on the lidar-scan data for accurate tracking. After the position data of people are extracted from the lidar-scan data using a background subtraction method, poses and behaviors of people, such as stopping, walking, and suddenly rushing out, are estimated based on a distributed interacting-multimodel (DIMM) estimator. The DIMM-based tracker works in any sensor network topologies, and therefore, this may provide a degree of scalability and robustness that cannot be achieved by conventional centralized interacting-multimodel (CIMM) based tracker with a central server. Experimental results reveal the tracking performance of the proposed DIMM-based tracker by comparison with the CIMM-based tracker.

This paper investigates human activity recognition based on biological information estimated by a non-contact measurement method. A facial image sequence of a person is utilized to estimate the biological information, such as heart rate, heart rate variability and blinking. Experiments that impose physical and mental tasks are conducted in order to gather the biological information, and then experiments of recognising the person's active state are conducted to evaluate three active states of resting, physical burden and mental burden. In order to classify person's active state, machine learning approaches, such as multilayer neural network and support vector machine, are used to design recognition systems and their characteristics are investigated. In person-independent recognition, neural network using all biological information features including dynamics information can attain a recognition rate of 82% while support vector machine can achieve the recognition rate of 97%. Experimental results demonstrate that using multimodal biological information is feasible for recognising the person's active state and that support vector machine is suited for this task.

This paper presents a method for recognizing the state of a user's brain activity by applying fuzzy reasoning to signals measured by a two-channel compact near-infrared spectroscopy (NIRS) device and investigates its feasibility and characteristics with regards to developing a compact brain-machine interface that utilizes NIRS. Distance-type fuzzy reasoning with membership functions of a singleton fuzzy number in the antecedent/consequence parts is applied to the oxyhaemoglobin concentration change obtained recorded by the compact NIRS device to recognize active and calm brain activity states. In the experiments in which subjects are assigned the tasks of reading silently to themselves, an average recognition rate of 64% can be achieved by the proposed method. As a practical application of the proposed method, experiments to control a robot arm using brain activity recognition result are conducted. Experimental results indicate that a recognition rate of 70% or higher can be achieved by utilizing a feedback interface that allows users to visually understand the state of their own brain activity even when a user is unfamiliar with the system.

This paper presents laser-based tracking of moving objects conducted by a group of mobile robots located near one another. Each robot finds moving objects such as people, cars, and bicycles in its own laser-scanned images using a binarized occupancy-grid-based method. It then sends laser measurements related to the detected moving objects to a central server. The central server estimates the pose and size of the moving objects via the Kalman filter based on received measurements; it then feeds that information back to the robots. Rule-based and global-nearest-neighbor-based data associations are applied for matching of tracked objects and laser measurements in multitarget environments. In this cooperative tracking method, the central server collects the laser measurements from all robots; hence, the robots can always track invisible or partially invisible objects. The experimental results for two robots in an outdoor environment validate our tracking method.

Energy efficiency of wireless clients is an important issue in wireless communications. When multiple network applications are running concurrently on a single wireless client, packets belonging to each application are sent and received independently, but these packets are multiplexed at the MAC level. This uncoordinated behavior makes it difficult to control sleep timing. In addition, transitioning between active and sleep modes frequently will consume a non-negligible amount of energy. To address these difficulties, we have proposed SCTP tunneling: a transport-layer approach that resolves the coordination problem and so reduces the energy consumed by multiple TCP flows on a wireless LAN (WLAN) client. In this study, we perform experiments with off-the-shelf WLAN devices to assess the energy efficiency and the transfer time of SCTP tunneling. We show experimentally that SCTP tunneling with unscheduled automatic power save delivery (U-APSD) can save energy when compared to U-APSD alone, while still increasing file transfer speed.

This paper evaluates human emotional change by sound stimuli focused on chord progression in jazz music and conducts computational emotion classification from physiological information. Psychological experiments using chord progression tunes as sound stimuli are conducted with 117 subjects and the result of subjective evaluation shows that positive emotional valance chord progression tunes that have ascending fourth aroused positive images, and negative emotional valence chord progression tunes that have chromatic descent aroused negative images. Psychophysical experiments using chord progression tunes to excite emotions in subjects are conducted to gather acceleration plethysmogram data. For computational emotion classification, multi-layer neural network using feature values extracted from heart rate and acceleration plethysmogram is used to discriminate emotional class. In experiments of computational emotion classification, an average of 38.3% classification rate is attained in three emotions - positive, negative, and neutral.

This paper presents a method for pedestrian tracking in urban environments using in-vehicle multilayer laser lidar (MLLR). The MLLR that we developed irradiates the laser in six scanning planes by a polygon mirror mechanism, and thus objects with height are observed with the plural scanning planes. MLLR outputs are modified by GPS data and are mapped onto a grid map. Pedestrians are found based on the occupancy grid method, and they are tracked via Kalman filter in conjunction with global nearest neighboring (GNN) based data association. A track management method improves tracking accuracy in real worlds. Our tracking algorithm works well in a low-performance computer environment. The experimental results in different scenarios such as intersection and on the community road validate the proposed method.

This paper presents a people tracker that uses a two-layered laser range sensor (LRS) and a fisheye camera. The LRS detects waists and knees of people, and the camera captures color images of torsos of people detected by the LRS. From the data of people's position obtained by the LRS and the color histogram obtained by the camera, heuristic-rule-based and global-nearest-neighbor-based data association can identify multiple people in crowded environments. The identified people are tracked via a model-based tracker; an interacting multiple model estimator is applied to track people maneuvers, such as walking, running, sudden starting/stopping, and sudden turning. Simulation and experimental results show that our people identification and tracking method provides better performance than conventional methods.

This paper presents a method for automated human detection using fisheye lens camera. We introduce a probabilistic model to describe the wide variation of human appearance in hemispherical image. In our method, a human is modeled as variable shape features of body silhouette and head-shoulder contour. These features are extracted from the human images taken at various distance and orientation with respect to the camera, and form the training data set for template modeling. Non linear template models are build by the combination of Principal Component Analysis (PCA) and Kernel Ridge Regression (KRR). Finally, the problem of human detection is formulated as maximum a posteriori (MAP) estimation using above model. Experiments are conducted on indoor space where a fisheye lens camera is installed on the ceiling. The robustness and accuracy of our method is discussed through the experimental results.

This paper describes a gesture recognition algorithm based on electrooculographic (EOG) signals measured by a simple brain-computer interface to estimate a user's intentions. A dynamic programming (DP)-matching-based gesture recognition algorithm is proposed and its characteristics are discussed regarding improving the eye movement recognition rate, along with the potential for control applications in a hands-free manipulation systems. Experimental results show the feasibility and effectiveness of the estimation algorithm using the DP-matching-based method. The control of a walkthrough in 3D virtual space is described as a practical example. As an extension of the EOG gesture recognition algorithm, the recognition method is applied to estimate multiple directions of eye movement by using multiple electrodes attached around the eyes, with experimental results showing the feasibility of the method.

This paper investigates a human body posture estimation method based on the back projection of human silhouette images extracted from multi-camera images. The multi-camera system is based on a server-client system with local network of 1000 Base-T to achieve a voxel 3D reconstruction of human body posture in real-time. In order to extract significant points of the human body such as head, neck, shoulders, elbow joints, hands, waist, knee joints, and toes in 3D, an articulated cylindrical human model is applied to the voxel reconstruction of human body. To evaluate the proposed human body posture estimation method, 3D reconstruction experiments of human body posture and extraction experiments of human bodypsilas significant points are carried out. The system runs in real time (9 frames/sec with 50 times 50 times 50 voxel resolution) and the experimental results confirm both the feasibility and effectiveness of the proposed system in 3D human body posture estimation.

This paper describes a method of fault diagnosis of internal and external sensors for mobile robot. As our previous works diagnosed hard and noise faults of internal sensors, in this paper, scale faults, in which sensor gains deviate from the fault-free condition, are handled for three internal sensors (two wheel-resolvers and one gyro). The sensor gains are estimated based on the robot velocity estimated with a fault-free external sensor (a forward-looking laser range sensor) via the scan matching method, and then the scale faults are diagnosed. Abrupt faults of the external sensor are handled in this paper. They are detected based on errors related to the scan matching; if the laser images in several successive scans are not matched, the external sensor is decided to be faulty. False detection of the external sensor faults in dynamical environments is improved based on the occupancy grid method. Experimental results in an indoor environment show the performance of our method.

This paper investigates a nonverbal interface system using bio-potential signals, such as EOG and EMG, measured by a simple brain-computer interface and discusses its possibility of controlling application in a hands-free manipulation system. A simple gesture recognition algorithm from EOG and EMG signals is presented to estimate the user's intentions and/or thoughts. To evaluate the effectiveness of the interface system, experiments of controlling a walkthrough application in 3D virtual space are carried out and the usefulness of the interface system is confirmed. As a practical application example, controlling an active camera with the interface system is also demonstrated.

No abstract available.

This paper proposes a nonverbal interface system using bio-potential signals, such as EOG and EMG, measured by a brain-computer interface and investigates its possibility of application to control of a hands-free manipulation system. A simple gesture recognition algorithm is presented to estimate the user’s intentions/thoughts from the EOG and EMG signals. To evaluate the feasibility and the characteristics of the nonverbal interface system, moving control experiments in 3D virtual space are carried out and the effectiveness of the proposed interface system is confirmed.

This paper proposes a nonverbal interface system using bio-potential signals, such as EOG and EMG, measured by a brain-computer interface and investigates its possibility of applying to control application. A simple gesture recognition algorithm is presented to estimate the user's intentions from the EOG and EMG signals, and the user's emotion is evaluated by using the EEG signal. To evaluate the feasibility and the characteristics of the nonverbal interface system, moving control experiments in 3D virtual space are carried out and the effectiveness of the proposed interface system is confirmed.

This paper proposes a human body posture estimation method based on back projection of human silhouette images extracted from multi-camera images. To extract human silhouettes from color images accurately, a real-time background subtraction method that utilized the brightness and chromaticity in RGB color system for each pixel of the color image is proposed. The threshold values for the brightness and the chromaticity are statistically defined in advance from the image sequence of a background. The pixel classification rules considering the brightness, chromaticity, and their threshold values, classify each pixel of an image observing a human with the background into the human silhouette or not. Using the obtained human silhouette images, a 3D human body posture reconstruction in real-time with silhouette volume intersection is also investigated. To evaluate the feasibility of our proposed human body posture estimation method, 3D reconstruction experiments of human body posture are carried out. The experimental system with a personal computer and three CCD cameras under single processing scheme is composed and the experimental results confirm both the feasibility and effectiveness of the proposed background subtraction method and the 3D reconstruction method in real-time.

A multi-model based approach to fault detection and diagnosis (FDD) of internal sensor for mobile robot is proposed. Three failure modes (hard failure mode, noise failure mode, and scale failure mode) of the sensor are handled; on the hard failure the sensor output is stuck at a constant value. The noise failure causes the sensor output with large noise. On the scale failure the gain (scale) of the sensor output differs from the normal. The detection and diagnosis of the hard/noise failure (HNFDD) is based on the variable structure interacting multiple-model (VSIMM) estimator; changes of the failure modes are modeled as switching from one mode to another in a probabilistic manner. The mode probabilities, which are estimated based on a bank of Kalman filters, provide the fault decision. The detection and diagnosis of the scale failure (SFDD) is achieved via single-model based Kalman filters, each of which is based on a model matching to a failure mode of particular sensors; the fault decision is made by comparing the model conditional estimates in the sensor gain. The proposed FDD algorithm is implemented on our skid-steered mobile robot with five internal sensors (four wheel-encoders and one yaw-rate gyro). Experimental results show the property of the FDD algorithm.

An interacting multiple-model (IMM) approach to sensor fault detection and identification (FDI) in the dead reckoning of mobile robots is proposed Changes of sensor normal/failure modes are explicitly modeled as switching from one mode to another in a probabilistic manner; mode probabilities and robot states are estimated via a bank of Kalman filters with mutual interaction. To provide better fault decision, mode probability averaging and heuristic decision-making rule are incorporated into the IMM based algorithm. The proposed FDI algorithm is implemented on our mobile robot. 16 system modes (one normal mode and 15 'hard' sensor-failure modes) of four internal sensors (two wheel-encoders, one steering potentiometer and one yaw-rate gyro) are handled. Experimental results validate that the proposed FDI algorithm allows robust navigation of the robot subject to sensor failures.

This paper proposes a control scheme to transport a palletized load by several wheeled vehicles (automated guided vehicles in factory). The planar Cartesian manipulator with two-degrees-of-freedom prismatic joint and one-degree-of-freedom rotary joint is mounted on each vehicle so that the vehicle with nonholonomic constraint and slow dynamic response can dexterously manipulate the pallet. Each vehicle-manipulator system independently controls the acceleration of its own hand grasping the pallet based on the prescribed mechanical impedance model. The kinematical redundancy of the vehicle-manipulator system is exploited to keep the manipulator arm within its workspace while executing the hand desired acceleration faithfully. The simulation result for the object-transportation by four vehicle-manipulator systems illustrates the validity of the proposed control scheme.

This paper proposes a coordinate dynamic control method for the transfer of a common heavy object by several wheeled mobile robots. The coupler, which comprises an active prismatic link and a passive rotary joint, is installed on each robot so that even an ordinary nonholomonic mobile robot with inferior mobility can correctly manipulate the object. A hierarchical coordinative control system composed of a leader and other staff robots is designed to synchronously control the motions of the robots under severe dynamic interactions. The leader robot successively defines the desired motion of the object along its reference trajectory and broadcasts the information to the staff robots. Each staff robot controls the motions of its own body and the onboard prismatic link so as to achieve the desired object behavior without sideways slippages. When the prismatic link approaches the limit for its movement, the staff robot requests the leader to temporarily modify the desired behavior of the object. The control algorithm is formulated on the basis of the kinematics and the dynamics of the wheeled mobile robot. Simulation results illustrate the validity of the method.

This paper proposes trajectory generation and tracking control methods for the advanced transportation system METROS (Multiple Transfer Robots System). The methods enable several transfer robots to transfer materials cooperatively in an automated factory. The material's reference trajectory guaranteeing smooth motions of robots is represented by simple combinations of line trajectories with constant velocity profiles and transition ones with 8th order polynomials. In order to transfer the material stably along the reference trajectory, a cooperative control architecture which consists of a master robot and slave ones is introduced. The master robot determines a behavior of the material for the specified reference trajectory, and thereafter distributes the forces required to realize the desired behavior to the slaves. Each slave robot generates its own trajectory and controls the speed and steering so as to exert the force commanded by the master robot. An impedance model is defined to determine the behavior of the material so that the material can behave as a desired mass-spring-dashpot system. The simulation results justify both the proposed methods.>

This paper proposes two passive reference methods of position finding for an automatic guidance system of a ground vehicle using laser beam and corner cube(retro-reflector). One is a positioning method for a vehicle guided along a flexible and arbitrary path. This system is composed of corner cubes set on pre-determined three reference points of ground and a laser beam scanned on a vehicle. Another is a positioning method for a vehicle (automobile) guided within a traffic lane on roadway, which is used for a compensations of the positional errors caused by a self- contained measuring system. This method is realized by a chain of corner cube pairs set on both sides of roadway and four laser beams installed on a vehicle. The experimental results using our prototype system and the simulation results are shown with some discussions about the feasibility of real system using our methods.