This paper proposes a people identification method based on the sitting patterns. This method uses weak evidences from pressure sensors, accelerometer sensors, and light sensors placed on a chair to recognize who is sitting on the chair without any psychological and physical burden on users. We discuss how we have implemented the system using softmax regression model, gradient descent algorithm and nearest neighbor search algorithm. Our experimental result shows that this method can be used in places which has private properties such as a home or small a office. Nowadays, there are several biometric people identification methods such as fingerprint based[1], iris based[2] or by the using of vein[3]. These biometric identifiers are strong and suitable for applications that request high accuracy such as security applications. However these identifiers annoy user with the requests for specific actions. For example, in the case of fingerprint, users have to properly touch a fingerprint scanner or in the case of retina, users have to look at retina scanner for a while, which might cause a psychological and physical burden on users. These methods also need delicate and expensive devices such as fingerprint scanner or retina scanner. In such situations as inside of a house or a small office with a small number of users, we do not need high accuracy as those available with strong identifiers. For example, in a small office, which employees come from some different country, an employee comes to the office, sits on a public chair and turns on a public computer. And then, a greeting sound of his/her country comes out of the speaker and that computers' language will be automatically change to his/her native language. Is it interesting? For the other scenario, an office has a meeting but the boss is in a business trip so he/she uses a robot for teleconference. The robot stands in the middle of meeting room and when the boss wants to talk with one of his/her employees, he/she only has to let the robot knows the employees' name instead of rotating robot by hand. Both of these scenarios can be realized with one of the above people identification methods, but using biometric identifiers for this scene is wasteful and unnecessary. Any mistake of people recognition in these scenes is not a big problem, users can easily overcome it by simple actions. So, it is acceptable to inference to the user who is sitting beforehand based on weaker evidences. Collecting weak evidences also can be implemented without any psychological and physical burden of users. This paper proposes an easy deployment and inexpensive people identification method that uses weak evidences from pressure sensors, accelerometer sensors and light sensor placed on a chair. The reason of using pressure sensor is the difference of weight among users. Also, we think that the sitting patterns are different between users so we use accelerometer sensors to recognize the movement of the chair when user sit in it. The light sensor is used to measure the coverage of user in the chair. We have used softmax regression model, a supervised learning algorithm and gradient descent algorithm, an algorithm to solve optimization problem to inference who is sitting in the chair Remainder of this paper is organized as follows. Section 2 describes the design and implementation of system. The softmax regression model, gradient descent algorithm, nearest neighbor search algorithm and how they are used are discussed in Section 3. Section 4 shows the result of our experiment while Section 5 is about related work. Conclusions and future work are described in Section 6.
2 2.1
in Figure 1(a) is a wireless sensor network mote which developed by Sun Microsystems. One SunSPOT device has three types of sensor including an accelerometer sensor, a light sensor and a temperature sensor. In this research, we only use one SunSPOT device to sense accelerometer and light data. These data can be sent to a computer for processing through a base station as shown in Figure 1(b). (a) SunSPOT device
(b) SunSPOT base station
for a pressure sensor and Figure 2 shows how it can be viewed in fact. We want to use as least as possible sensors to reduce the cost of the system and we think that four is a good number. It is enough for people recognition issue based on weak evidences. The software diagram of this system is shown in Figure 4 . The data receiver module receives data from sensors and forwards it to the data processing module. In here, the data is normalized to be used in the learning module. In the learning module, the system uses softmax regression model and gradient descent algorithm to inference to the user who are sitting beforehand, output result and get confirmation from user through the user interaction module.
of users as a classification problem. The system classifies the data from sensors into groups. A group represents a user so the number of groups equal to the number of users exists in the data training set.
system will determine which group that this data set belongs to. By this way, the system can recognize the user who is sitting beforehand.
fication problem. The ”weight” used in the nearest neighbor search
algorithm are determined by Softmax Regression Model and
Gradient Descent Algorithm. The softmax regression model is discussed
in subsection below.
Softmax Regression Model[
hθ(x) takes the form: 2P (y(i) = 1| x(i); θ)3 6P (y(i) = 2| x(i); θ)7 hθ(x(i)) = 664 ... 775 =
P (y(i) = k| x(i); θ) 2eθ1T x(i) 3 Pjk=1 1eθjT x(i) 66466eθ2T...x(i) 7777 eθkT x(i) 5
Here, m dimensional vectors θ1, θ2, . . . , θk ∈ <m are the parameters of this model and θiT is the transpose vector of θi. Notice that the term Pjk=1 e1θjT x(i) normalizes the distribution, so that it sums to one. For convenience, we will also write θ to denote all the parameters of our mode.
given input vector x belong to. That is the class label that have maximum estimated probability. But the hθ function is expressed by the θ parameters, so we have to find all the θ parameters.
J (θ) = − n1 » Xn Xk Q(i, j) log i=1 j=1 eθjT x(i) l=1 eθlT x(i) Pk –
Q(i, j) = 1 0 if y(i) = j otherwise P (y(i) = j| x(i); θ) = eθjT x(i) l=1 eθlT x(i) Pk
the cost function. There are several methods to do it such as gradient descent algorithm or limited-memory BFGS algorithm. In this paper, we used gradient descent algorithm that is described below. 3.2
to minimize the cost function J (θ). It starts with some ”initial guess” for θ, and that repeatedly change θ to make J (θ) smaller, until hopefully converge to a value of θ that minimizes J (θ). The gradient descent algorithm repeatedly performs the update: θj := θj − α
J (θ) ∂ ∂θj
of softmax regression model, we need to compute the partial derivative of cost function J (θ). It is shown by the equation below.
i=1 ∇θj J (θ) = − n1 Xn »x(i)(Q(i, j) − P (y(i) = j| x(i); θ)) –
In particular, ∇θj J (θ) is itself a m dimensional vector, so that its l-th element is ∂θ∂jl J (θ), the partial derivative of J (θ) with respect to the l-th element of θj . So we can use it to compute the update value of all parameters in softmax regression model.
subtract fixed vector ψ from it, so that every θj is now replaced with θj − ψ ( for every j = 1, 2, . . . , k), we have:
P (y(i) = j| x(i); θ) =
e(θj −ψ)T x(i) Pk
l=1 e(θl−ψ)T x(i) = eθjT x(i) e−ψT x(i) l=1 eθlT x(i) e−ψT x(i) = Pk eθjT x(i) l=1 eθlT x(i) Pk
of parameters (θ1, θ2, . . . , θk), then it is also minimized by (θ1 − ψ, θ2 − ψ, . . . , θk − ψ) for any value of ψ. Thus, the minimizer of
adding a weight decay term λ2 Pik=1 Pjm=1 θi2j which penalizes large values of the parameters. Our cost function is now J (θ) = − n1 » Xn Xk Q(i, j) log
imize solution. Also, the gradient descent algorithm is guaranteed to converge to the global minimum. To apply the gradient descent algorithm, we also need the partial derivative of this new definition of
i=1 ∇θj J (θ) = − n1 Xn»x(i)(Q(i, j) − P (y(i) = j| x(i); θ)) + λθj –
mize the cost function J (θ), we will have a working implementation of softmax regression model. 3.3
Nearest neighbor search (NNS)[
n CP (q) = { p| p ∈ S; dpq = min dSiq} i=1
We use one accelerometer sensor, one light sensor and four pressure sensors placed on a chair for people recognition, so we have eight values of sensor data. • • • • • • • •
r = (Ax, Ay, Az, Light, A1, A2, A3, A4) will be created. This array is called RA and it describes the information of one sitting time of a user. In our data training set, there are 10
RA = { r1, r2, . . . , r15}
We use nearest neighbor search with 8-dimensional Euclidean space for this classification problem. But the Euclidean distance function we use has a litter different to general function. Because these eight sensor data affect to result in different ways, we modify the Euclidean distance function like this: v u 8 dpq = tuX θi(qi − pi)2
i=1
we use softmax regression model and gradient descent algorithm to determine them. Our people identify process can be described as following:
age of all records of this RA and the average of all records of all RA in data training set and use softmax regression model to compute the parameters used in nearest neighbor search algorithm.The gradient descent algorithm is implemented with learning rate α = 0.001 and λ = 0.001 to minimize the cost function in softmax regression model. Finally, we use nearest neighbor algorithm with determined parameters to classify the new RA to one of k class labeled. The result is the user whom this class labeled stand for. There are always 10 RA for one user in our data training set, but the data training set is dynamic. When a user sit down to the chair, after the system receives the confirmation from the user, in data training set, the oldest RA of this user is replaced by the newest RA. By this way, the system can adapt with the change of users’ sitting pattern. 4
with a group of five people and in the other case, we evaluated with a group of ten people. In both case, one person must sit in a chair twenty times, ten for training and ten for testing. Figure 5 shows the result of first case while Figure 6 shows the result of last one.
tified with the accuracy as 100%. One people with the accuracy as 90%, it means that there was only one mistake.
this method certainly can be used in a small office or inside a house with a small number of users. 5
Masafumi Yamada et al.[
We have proposed a people identification method based on sitting patterns of user. This method used weak evidences collected by accelerometer sensor, light sensor, and pressure sensor placed on a chair to inference who is sitting on it. We considered this problem as a classification problem and use nearest neighbor search algorithm with ”weight” to resolve it. The ”weight” used in nearest neighbor search algorithm is determined by softmax regression model while the cost function is minimized by gradient descent algorithm. We also presented the result of experiments which shown that this people identification has the accuracy enough to be used in places which have private properties such as inside of a house or a small office.
sensor to a chair are the things to be discussed in the future. Moreover, the evolution of performance increasing the number of people need to be studied. We also intend to implement a module to recognize the posture of user or the users’ mood.