Masahiro Mochizuki

A pedestrian flow analysis system using Wi-Fi packet sensors to a real environment

The authors have been developing the system, which analyzes pedestrian flow using Wi-Fi packet sensors. The sensors collect Wi-Fi packet called probe request packet, which is transmitted from a smartphone to search Wi-Fi access points. In addition, the cloud storage server is running to manage observed packets centrally and to compute pedestrian flow in real time. Additionally, user movement history is vitally important and we have to pay close attention to handling that kind of data. Therefore, the system runs with an anonymization method and a cryptographic function. Some kinds of demonstration experiments were held in real environment. As a result, it was confirmed that we can analyze the rough tendency of pedestrian flow using the present system and simple analysis methods.

User activity recognition method based on atmospheric pressure sensing

Several studies have been conducted on context recognition as well as hobby and preference extraction by analyzing the data obtained from the sensors in a smartphone. As a smartphone component, a barometer is expected to be useful for activity recognition because of its low power consumption. In this work, we propose an activity recognition method of classifying a users state into indoor and outdoor states and using a barometer at each state. In the proposed method, the floor of a building on which a user is located is estimated by determining atmospheric pressure variations sensed in the indoor state, and the users location is estimated by determining atmospheric pressure variations according to the user movement along a track in the outdoor state. In particular, this paper delineates the method of estimating the current floor on which the user is located. We confirmed that it is possible to closely estimate the current floor of the building in the case of user movement among eighteen floors.

A recognition method for continuous gestures with an accelerometer

Along with the spread of smart phones and wearable devices, systems that recognizes gestures such as punch and chop using an accelerometer have recently been attracting a great deal of attention. However, these systems do not consider the situation that gestures are performed continuously from/to other gestures or untrained movements and can not recognize such gestures accurately. This paper proposes a method that recognizes gestures performed continuously without explicit intervals. The proposed method detects segments similar to template data of target gestures and chooses the most likely segment. In order to evaluate the effectiveness of the proposed method, an experiment is conducted with five subjects. The subjects conducted three types of gestures drawing a graphic symbol in the air; circle, triangle, and cross in five kinds of situations, and ten types of gestures drawing a numeric character in the air; zero to nine. The average F-measure of graphic symbol achieved 0.78 and the average F-measure of numerical character achieved 0.79.

Cross-assistive approach for PDR and Wi-Fi positioning

In indoor positioning using Wi-Fi, there is a problem that the accuracy is not stable by the occurrence of large errors. Large errors tend to occur when density of wireless LAN access points is low or the radio wave condition is unstable. Furthermore, as for positioning utilising smartphone, it takes a while to scan Wi-Fi beacons. Thereby, errors tend to occur while user is moving. Because it is difficult to observe exactly Wi-Fi beacons. Accordingly, the authors proposed Cross-Assistive Approach for PDR and Wi-Fi Positioning. First of all, fingerprinting that is often used Wi-Fi positioning is improved by confining fingerprints to location where is estimated by PDR. As a result, this approach improved the accuracy about 2 meters. Furthermore, in order to correct accumulated errors in PDR, the authors proposed a method that corrects PDR with accurate Wi-Fi positioning results. Additionally, the authors proposed a method that estimates the accuracy of Wi-Fi positioning results. The mean error of accurate Wi-Fi results estimated by the accuracy estimating method was 0.98 meters. Thus, the accuracy estimating method detected accurate Wi-Fi positioning results effectively. In the comprehensive evaluation, our approach improved an existing Wi-Fi method about 3.4 meters by assisted PDR with Wi-Fi positioning and assisted Wi-Fi positioning with PDR cooperatively. Moreover, this approach enabled accumulated errors in PDR to be corrected.

User Attribute Classification Method Based on Trajectory Patterns with Active Scanning Devices

Technologies for grasping the distribution and flow of people are required for urban planning, traffic planning, evacuation, rescue activities in case of disaster, and marketing. In order to grasp what kind of attribute the distribution and flow of people are formed, this paper proposes a method that estimates the attributes of users. As a method of estimating user attributes, we utilize probe request frame of Wi-Fi that smartphones are emitting. Probe request frame includes MAC address, enabling us to acquire the movement trajectory of a user by tracking the MAC address. By using the feature values obtained from the movement trajectory of the user, users are roughly classified into several types. In this paper, we focus on the user attribute estimation in underground city comprising of stations, shops, restaurants and so on. Through the practical experiment at Osaka underground city, we confirmed that the proposed method can classify the users into commuter or not by using the intervals between probe request frames.

GERMIC: Application of Gesture Recognition Model with Interactive Correction to Manual Grading Tasks

Gesture-based recognition is one of the most intuitive methods for inputting information and is not subject to cumbersome operations. Recognition is performed on human’s consecutive motion without reference to retrial or alternation by user. We propose a gesture recognition model with a mechanism for correcting recognition errors that operates interactively and is practical. We applied the model to a setting involving a manual grading task in order to verify its effectiveness. Our system, named GERMIC, consists of two major modules, namely, handwritten recognition and interactive correction. Recognition is materialized with image feature extraction and convolutional neural network. A mechanism for interactive correction is called on-demand by a user-based trigger. GERMIC monitors, track, and stores information on the user’s grading task and generates output based on the recognition information collected. In contrast to conventional grading done manually, GERMIC significantly shortens the total time for completing the task by 24.7% and demonstrates the effectiveness of the model with interactive correction in two real world user environments.

Hybrid approach for reliable floor recognition method

Location-based services and context-aware services of smartphones have been paid much attentions. For these services, location information is important. Therefore, several positioning methods have been developed. This research focuses on unnecessary operation of GPS in seamless Indoor/Outdoor positioning. In order to reduce unnecessary behavior of GPS, we propose two floor recognition methods. First method is an advanced floor recognition method by using Wi-Fi and barometer in smartphones, when we know the location of Wi-Fi APs and the height between floors of buildings. Second method is to learn and infer if a floor of a building whose location of APs is unknown is connected to outside of the building. We conducted an evaluation of the proposed methods. As a result, the accuracy of the first method was 96%. The evaluation result also shows that the second method can learn and infer the floor which is connected to outside of the building.

Detecting aged deterioration of a radio base station map for wi-fi positioning

Currently GPS is utilized for localization but is not reliable indoors or underground where its signals cannot reach. Instead, localization using Wi-Fi is investigated in the field of indoor positioning. Wi-Fi localization is performed using a radio model created by observed information collected beforehand; however, the model deteriorates over-time as Radio Base Stations (RBS) appear, disappear, and are relocated. As a result, localization accuracy decreases. Accuracy of positioning is affected by RBS and the number of RBS used for localization, thus, updating the model is required whenever deterioration occurs. There is no way to detect deterioration and necessity to update. This research investigates a method to detect such deterioration only by analyzing logs collected from navigation applications. Appearance and disappearance of RBS can be detected by analyzing observed date with its interval related to its average and standard deviation. Relocation of RBS can be detected by finding change-point of RBSs co-occurrence with other RBS. As results, it is demonstrated that RBS which possibly appears and/or disappears can be specified, if observed over certain days. It is also demonstrated that RBS which possibly be relocated can be specified by finding the best threshold.

Adapting Wi-Fi samples to environmental changes automatically

In recent years, a positioning method which utilizes wireless LAN without using GPS has attracted attention. Especially, in the case of a method which combines absolute position with a Wi-Fi radio environment in advance, the cost of operation and management becomes enormous. Therefore, by sampling Wi-Fi radio information observed at points where users stay frequently or in the long-term, a method which automates to collect and update the Wi-Fi radio information has been proposed. In the case of a long-term operating, the positioning accuracy, however, decreases because this method does not perform well in maintaining and managing samples. It cannot adapt samples to environmental changes although Wi-Fi radio signals change in case of long-term operating. Accordingly, this paper proposes a new calculation formula for improving a positioning accuracy. The formula is calculated with the weight of each base station for avoidance of ill-behaving stations. In addition, this paper also proposes the automated management system with two steps. It adapts samples to changes of Wi-Fi radio signals and a users behavior. As a result, a positioning accuracy of the new system is higher than existing one.