Abstract— library and we use logistic regression algorithm

Abstract— Intelligent Assistants are
becoming an essential part of our life they are present in smartphones like
iPhone has Siri Android has Google Assistant and now they are becoming a core
part of computer operating systems i.e. Windows has Cortana, Linux has Stella. These
Intelligent Assistants are aimed to make life easier by helping the user in
routine tasks like a Human Personal Assistant do. But they are unable to take
place of Human Personal Assistants, one of the main reason behind this is that
these Intelligent Assistants cannot understand Human emotions. In this paper we
have proposed a method that will make these Intelligent Assistants capable of
understanding Human Emotions. In the proposed method we are using open source
Intelligent Assistant “Open Assistant”, as the user calls the Assistant a real-time
picture of the user will be taken using OpenCV library and we use logistic
regression algorithm to train on a provided dataset and evaluate the new image.
It is detected whether the user is smiling or not then this information is
passed to Open Assistant to make him able to react according to user’s emotion.

1.     
Introduction

In present day Intelligent Assistants
are becoming a core part of our life. These personal assistants can remind you
about your daily tasks, they can send emails, even they can order lunch for
you. These Intelligent Assistant are equipped with a bit of artificial
intelligence (AI), they can understand human language, respond to general
queries and their artificial intelligence is also used to ensure that their
responses are in line with the expectations of the user. But these Assistants
are lacking the understanding of human emotions. In 1968, Albert Mehrabian 1
pointed out that in human to human interaction 7% of communication is
contributed by verbal cues, 38% is contributed by vocal cues and major portion
55% is contributed by facial expressions. And these facial expressions are used
to understand emotions. If a machine can identify human emotions, it can
understand human behavior better, thus improving the task efficiency 2. A lot
of work has been done for making these intelligent assistants efficient but
techniques for making them capable of understanding human emotions are yet not
available.

In this paper, we have proposed a
technique that will make these intelligent Assistants capable of understanding
Human emotions. For this we are using an open source Intelligent Assistant
“Open Assistant” as the user call the Open Assistant the webcam will take
user’s picture then OpenCV computer vision library and logistic
regression algorithm would be used to detect whether the user is smiling or neutral
then this information is forwarded to Open Assistant for executing particular
code. This technique can be used for any Intelligent Assistant to make it more
efficient and humanoid.

2.     
Related Work

The technology giants like Google,
Microsoft are working for making their Intelligent Assistant humanoid for
bettering the tasks efficiency. There work is not available publicly but the
general perspective about their work for Intelligent Assistants is that, they
are using data stored on user devices and their daily device usage patterns for
making Intelligent Assistant responses inline with the expectations of user. At
the time of writing this paper no research work was found for making the
Intelligent Assistant capable of understanding human emotions.      

3.     
Methodology

The following is the algorithm
for making Intelligent Assistant capable of understanding human-emotions:

Fig: Block Diagram
for emotion recognition system

a) OpenCV

The Intelligent Assistant starts
by a keyword spoken by the user, as the user speaks the keyword a picture of the
user with either a smiling or neutral facial expression using a webcam is
taken, this is done before the Intelligent Assistant done any processing. Then
our program uses an algorithm adopted from the OpenCV library to localize the
mouth area.

b) Vectorization

We resize the image such that the
output is 28 pixels by 10 pixels image only containing the person’s mouth and
surrounding areas. The images are converted to grayscale and then flattened
into a vector of length 280, with each entry representing the grayscale of a
pixel.

c) Logistic Regression

We built a logistic regression
program that will take the user-­provided image vector and determine whether
that person was smiling or not. First, the logistic regression is built to take
an input of dimension 280. It applies a set of weights to that input and then
yields a single scalar. Whether the activation is closer to 0 or 1 determines
whether the model will say that the original person was smiling or not.

Before the logistic regression can
classify the user­-provided image, we trained the model using gradient descent.
We used 64 neutral images and 43 smiling images from online datasets to train
the model and fine­tune the weights. With the suitable weights and biases, we
can input the user’s processed mouth image into the model and the network can
predict whether that person was smiling or not.

d) Open Assistant

The
different responses of Open Assistant are classified according to the emotion
of the user like one portions of code for neutral and one portion for happy,
when the smile is detected the code of Open Assistant for happy user is
executed.

 

4.     
Conclusion

In this paper we have used OpenCV
computer vision library, Logistic Regression program and Open Assistant to propose
a method that can provide emotion understanding to Intelligent Assistant which
are becoming an essential part of our lives.

In the future work more emotions
can be detected giving capability to Intelligent Assistant to react more
effectively.

REFERENCES

 1
Mehrabian, A. Communication without words, Psychology Today, volume 2, pp
52-55, 1968.

2 Suchitra, Suja P and Shika Taripathi,
Real-Time Emotion Recognition from Facial Images   using Raspberry Pi II, 2016 3rd
International Conference on Signal Processing and Integrated Networks (SPIN) 

BACK TO TOP