Muneyuki Sakata
GPS (Global Positioning System) is a powerful location method, but the public network services like ``Hotspot'', for instance, are implemented within the buildings where out of the reach of GPS in many cases. For indoors location, there are also some methods to acquire people's position or behavior in a room. Most of the methods using color images from a camera have a limit in the stability of position detection or user extraction by using matching processing of a color. Furthermore, each user is required to do an explicit log-in operation, for example typing their names and passwords, so that the system starts to aware and identify he/she. And the method using the combination of the transmitter and receiver of an electric wave or an ultrasonic wave, it is necessary to install many sensors, while highly precise position detection is possible. We propose a new framework of the location system to locate and identify the mobile PC users without "log-in operations to be watched", based on a ubiquitous mobile computing environment. We introduce a combination of the wide-angle camera and the IR(InfraRed)-tags which emit IR light. IR-tags are connected to the system through the wireless network and the lighting patters of the tags can be controlled by the system. Our approach is to employ the wide-angle camera for both spatial and temporal process. The spatial process traces the multiple users' positions and the temporal process identifies each user and associates with his/her position.
In this document, we describe the details about the process flow in the proposed system, ALTAIR in next section.
Figure 1:ALTAIR
The Location Server is on the top level and manages the location information for the all users in the whole area. According to the demands from the other application systems, information is sent to an application server which performs location related services. Local Location Manager(LLM) locates and integrates the users' positions in a section or a room, where users are carrying the mobile PCs. A LLM always communicates several Tag Detectors(TD). A TD captures the image sequence from a wide-angle camera to detect the lights from the IR-tags. The positions and lighting patterns of the IR-tags are transmitted to the LLM. Several camera systems of TDs are installed in each room. LLM manages and renews a list of the users' entry, according to the communications to the DHCP server and the TDs. To identify the users, LLM sends massages to a IR-tag to blink in order to associates the blinking position with corresponding IP address.
IR-tag is a small-sized marker with a IR LED which is controlled by the mobile PC though a serial connection, RS232C. The lighting pattern of the IR LED on the tag is changed according to the massages that the PC receives via network from the LLM.
Figure 2:IR-tag
A LLM merges and integrates the information sent from the TDs that cover an assigned section or a room and update the user map. The information from multiple TDs are buffered and merged by a LLM. Comparing the buffered data with the latest map, tracking of the each user's position is performed by nearest neighbor search. A LLM has a queue which contains all the users entry in the section expressed by their IDs.
LLM sends a request message to a user's IR-tag to blink in a specified manner. The user to sent the message is chosen according to the queue. If the blinking LED is detected by the TDs, LLM can identify the user and successfully renew the location. Even in the case after a user is completely lost from the TDs because of visual occlusion, for example, the location can be recovered by this ``roll call'' strategy.
Figure 3: Flow of LLM
When a new user comes into a system area, the user's Mobile PC acquires the information for connecting with LLM which manages the area, and with network environment, such as IP address using DHCP service etc. Based on the information acquired, a user's Mobile PC connect to LLM. LLM receives the connection demand from a user, attaches ID to a newuser, and push the new user's ID to blinking queue. A new user is distinguished preferentially and tracking is started.
As shown in Figure 4, we settled target area is 6m-square. Two IR wide-angle cameras(WATEC 202D, 640x480 pixels) with fish-eye lenses were installed in the area so that they face each other. The height of the cameras are about 2.5m and depression angles are about 45 deg. In the target area, an LLM and two TDs were implemented with the PCs. We performed manual calibration for camera setup such as positioning, settling the center and the range of the camera images. IR LEDs are used for the Tags. The blinking pattern of the IR-tag was settled as the repetition of 50ms on and 50ms off, or 10Hz. The height from the floor of IR tag was assumed to be 1.2m for calculation of the positions. The mobile PCs are carried so that LED might always turn to the direction of one of the cameras.
図4: Setup for Experiment
図5: Tracking result from Tag Detector (left) と LLM (right)
Figure 6 shows the output of the LLM and unprocessed camera image for comparison. Each TD captured images with the speed of 18-22fps and stable count of the blinking required the image sequence of about 13 frames. Updating rate of the users' position map on the LLM was 2 fps and identification time for one person was 4 seconds. Even sparse setup using only two cameras achieved real-time tracking for multiple users by the proposed approach. Tracing rate is not fast, but ``roll call'' strategy ensures the identification of the individual IR-tags and we consider that the performance of the system is efficient for the office workers location. In the Figure 7, two users walk across changing the directions. Visual occlusion may occur and tracking error to exchange the users may happen but, eventually our system performed correct tracking and mapping.
On the other hand, some problems were found in the system setup. We need to establish systematic method to calibrate the camera systems otherwise the measurement error of the position was up to 1m at maximum. The system also suffered from frequent occlusion because of the directivity of IR LEDs. We are going to improve the IR-tag emission, designing the arrangement of the multiple LEDs.e fixed in this experiment. We performed manual calibration for camera setup such as positioning, settling the center and the range of the camera images. IR LEDs are used for the Tags. The blinking pattern of the IR-tag was settled as the repetition of 50ms on and 50ms off, or 10Hz. The height from the floor of IR tag was assumed to be 1.2m for calculation of the positions. The mobile PCs are carried so that LED might always turn to the direction of one of the cameras.
Movie mpeg(729KB)
FLASH(450KB)
図6: Output of LLM(Left), Image for comparison(Right)
We have proposed ALTAIR system and implemented the core part to identify and grasp positions of the mobile PC users using IR-tag and wide-angle cameras. As the experimental results show the ALTAIR system is capable of managing information of two or more users in real-time process and recover from the tracking error because of visual occlusion and so on. Our approach is practical and verifiable. Our future work is to implement the system to connect more sections to cover larger area such as the whole building and activate an organized application in an office environment.