Introduction:
This paper presents an implementation of a people centric sensing application called "CenceMe". This is a kind of application which can infer people's presence by using mobile phone's sensors. And it can analysis different sensor's data information to get a personal status then can be integrated into Social Networks such as Facebook. It has a split architecture optimized for performance and battery life of mobile phones. And has a user study used for initial testing and optimization of performance. The motivation of this project is that along with the increasing number of smart phone users and lots of different sensors are used in these smart phones, we can get people's location , status like running , walking, or sitting, etc, by collecting data through sensors. And then users can share their status on Social Networks, so people can get almost real time status from each other.
How it works?
The system can be divided into two parts, the phone software and the backend software on the server. It has a split level classification and a power aware duty cycle. On the phone side , it has 5 sensors. 1.the bluetooth sensor , it is used to sense the presence of other users and check the MAC address of the phone. 2.the accelerometer sensor, it is used to sense motion of users. 3.microphone, it is used to detect conversation or sound around users. 4.photo capture, it is used for image upload to server. 5. GPS sensor, it is used to determine location and vehicle travel. All the sensors will do sensing then the phone software will collect sensors' data and make classification to produce primitives. The phone software will find the right separation of what processing can be performed on the phone and what makes sense to perform offline on the backend server. The simple classifiers make sense to perform in the phone reducing the amount of sensor data that must be pushed to the backend. It also implement a power aware duty cycle mode to improve battery life without a significant hit in the responsiveness of the system. And the customized tag make it more humanity. The phone software has communication between the phone and server through wireless network. It has a GUI and a thing called ClickStatus to visualization the presentation of people's presence on the phone. And can upload of primitives to backend server.
On the backend software side, it based on Apache and Tomcat servers to interface the mobile devices. Data exchange is accomplished through the Apache XMLRPC library. All messages are driven by the mobile device. This includes sensor data and occasional ping messages. MySQL is used as the database tool to maintain the sensor data and other primitives. The server has more complex and more power classifier to the split level classification architectural. Basically it has a conversation classifier to build upon the audio primitives. A social context by using multiple sensor data to identify the context of the scene. A mobility mode detector to detect if user is travelling in a vehicle or not. A location classifier to detect the location of a user and a "Am I Hot" classifier attempts to map previous user events to a stereotype.
Personal opinion:
I think the key component of this system is users. The application get the basic resource from users , manipulate it and perform it by data sharing in the social network. The sensors in the mobile phone can provide an appropriate set of data features to build accurate models. These with access to internet increasingly incorporate personal information exchange into their daily routines via technologies. Then a more richly textured user experience can be provided by integrating automatically harvested, processed and shared sensor data into the mix. It is an idea of global sharing in the digital world. Actually , nowadays the digital world represent one of the new frontiers in online entertainment and business service. People lead virtual lives in these alternative worlds using personal avatars (such as WOW). We can use the sensors embedded in commercial mobile phones to infer real world activities , which we can in turn reproduce in public virtual environments, sharing with the virtual world.
Because such kind of social network is based on mobile phone sensors , a lot of private information will be uncovered. It may accidently tell other people the specify location and time of a certain user. And these location information will be upload and store in the database , it may have the chance to be used by other people. Therefore, user's behavior , activity, time schedule, even health condition and many other information may used by other people. So , the security problem is a very important aspect in design such kind of software. And because the sensing work is assigned to different sensors, they have different type of data format, if the user only wants to share part of these data, the integrity of information will be lack of reliability, because it can be faked. And I think the virus in the internet is also a threaten to this kind of application, because the basic idea is sharing data with people. So the virus may be shared from user to user. Another issue is how to handle the delay due to the activity recognition process on the phone, the transmission of the inferred activity to the backend , the time retrieval data from the database and display on the screen. Alone with the growing user data, it is hard to search certain data in the database, so it may need an index or any other method to increase the speed of retrieval. And also a shorter time of classification for primitives and facts is needed because many of the participants believe that real time access to buddies' sensing presence should be one of the features of the system.
In a word, CenceMe does a pretty accurate job of organizing the social events associated with users. The implementation make an ideal solution for a split mobile device and backend system. But it still exist pitfalls such as the reliability of sensing data from the sensors, the privacy and secure issue, better data transport and retrieval , shorter time of classification.
No comments:
Post a Comment