DeepSafe A Hybrid Kitchen Safety Guarding System With Stove Fire Recognition Based On The Internet of Things

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

IEEE INFOCOM 2021 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS) | 978-1-6654-0443-3/21/$31.00 ©2021 IEEE | DOI: 10.1109/INFOCOMWKSHPS51825.2021.

9484547 IEEE INFOCOM 2021 Demo

DeepSafe: A Hybrid Kitchen Safety Guarding System with


Stove Fire Recognition Based on the Internet of Things
Lien-Wu Chen and Hsing-Fu Tseng
Department of Information Engineering and Computer Science, Feng Chia University, Taichung 407, Taiwan
Email: lwuchen@fcu.edu.tw

Abstract—This paper designs and implements a deep learning dedicated webpage and smartphone APP of DeepSafe, kitchen
based hybrid kitchen safety guarding system, called DeepSafe, owners can instantly monitor the current status of the kitchen
using embedded devices and onboard sensors to detect abnormal and manually shut down the gas source as necessary.
events and block gas sources in time through the Internet of
Things (IoT). In the sensing mode, the DeepSafe system can According to the relevant literature we studied, DeepSafe
prevent the kitchen from fire/explosion disasters by detecting is the first system for hybrid kitchen safety guarding that
gas concentration, recognizing fire intensity, and estimating provides the following features: (1) DeepSafe integrates the
vibration levels. In the control mode, the DeepSafe system can gas sensor to properly detect gas leakage and employs the 3-
automatically block the gas source as detecting an abnormal axis accelerometer to instantly determine the vibration level
event, remotely monitor the kitchen status via real-time streaming
videos, and manually turn off the gas source using a smartphone for avoiding fire/explosion disasters. (2) DeepSafe adopts the
as necessary. To accurately recognize the intensity of stove fire developed deep learning based methods using conventional
and detect abnormal fire intensity, deep learning based fire and densely connected convolutional neural networks to accu-
recognition methods using conventional and densely connected rately recognize abnormal fire. (3) DeepSafe provides remote
convolutional neural networks are developed to further improve surveillance and control functions to owners for monitoring
the recognition accuracy of DeepSafe. In particular, the prototype
consisting of an Android based APP and a Raspberry Pi the kitchen environment and controlling the gas source any-
based IoT device with the gas detector, image sensor, and 3- time and anywhere. In particular, an Android-based prototype
axis accelermeter are implemented to verify the feasibility and consisting of the IoT device, diverse sensors, dedicated server,
correctness of our DeepSafe system. and smartphones is implemented to verify the feasibility and
Keywords: Automatic Gas Blocking; Deep Learning; Inter- correctness of our DeepSafe system.
net of Things; Mobile Device; Smart Kitchen. II. S YSTEM D ESIGN
I. I NTRODUCTION Fig. 1a shows the system architecture of DeepSafe. On
The rapid development of wireless communications and the kitchen side, an IoT device consisting of the Raspberry
MEMS technologies has made the Internet of Things (IoT) Pi development board, gas detector, image sensor, 3-axis ac-
pervasive, which can explore the sensing capability of IoT celerometer, and servo motor is installed between the gas stove
devices and the Internet connectivity among IoT devices to and gas pipeline to periodically report sensing data, detect gas
enable smart environments [1], such as smart house, smart concentration, recognize fire intensity, and estimate vibration
building, smart community, smart city, etc. In addition, the levels. Once detecting any abnormal event (i.e., excessive gas
connected mobile devices (e.g., smartphones or tablets) can concentration, large fire intensity, or high vibration level),
be used to monitor environmental information and perform DeepSafe automatically turns off the gas source through the
necessary control through remote IoT devices. In particular, servomotor and instantly sends the warning message of the
the kitchen is an important area of the house, where family detected event to the group of registered kitchen owners.
members can cook meals and store ingredients. Based on On the owner side, a smartphone/tablet is used to remotely
IoT technologies, recent studies for smart kitchens to detect monitor the current status of the kitchen, instantaneously
emergency events have been proposed using human activity receive the warning messages of the installed IoT device, and
recognition [2], gas concentration sensing [3], [4], [5], [6], manually turn off the gas source as necessary (before the IoT
and house fire detection [4], [5], [6]. device detects an abnormal event in the kitchen).
In this work, we design and implement a hybrid kitchen Fig. 1b shows the operation flow of DeepSafe for abnormal
safety guarding system, called DeepSafe, based on the IoT. event detection, automatic gas blocking, and instant warning
DeepSafe can provide cooking safety in the kitchen as using notification. The gas stove to be monitored is equipped with
stove fire. First, we use a gas sensor to detect gas leakage the IoT device of DeepSafe (which is in the sensing mode
in a real-time manner. Second, we design deep learning by default). In the sensing mode, the gas concentration Cgas ,
based methods to recognize abnormal fire through an image fire intensity Ifire , and vibration level Lvib of the kitchen
sensor. Third, we employ a 3-axis accelerometer to determine environment are kept sensing and recognizing through gas
vibration levels. The gas source can be automatically closed as parts per million detected by the gas detector, streaming video
gas leakage, abnormal fire, or great earthquake occurs. Finally, captured by the image sensor, and acceleration information
the sensing data and surveillance videos of the kitchen are obtained from the 3-axis accelerometer. There are three phases
immediately transmitted to the DeepSafe server. Through the in the recognition process of stove fire intensity that includes

978-1-6654-0443-3/21/$31.00 ©2021 IEEE


Authorized licensed use limited to: Chaitanya Bharathi Institute of Tech - HYDERABAD. Downloaded on February 09,2023 at 08:45:00 UTC from IEEE Xplore. Restrictions apply.
IEEE INFOCOM 2021 Demo

(a) (b) (c)

Fig. 1. (a) System architecture, (b) operation flow, and (c) system implementation of DeepSafe.

fire image classification (i.e., labeling), feature map extraction Message Queuing Telemetry Transport (MQTT [10]) protocol
(i.e., training), and intensity similarity determination (i.e., is adopted to collect and distribute the sensing data of current
recognizing). gas concentration, fire intensity, and vibration levels for re-
To accurately recognize stove fire intensity, we develop deep mote monitoring and control using smartphones, as shown in
learning based fire recognition methods using conventional and Fig. 1c(c). Once detecting an abnormal event (e.g., gas leak-
densely connected convolutional neural networks to further age), the warning message of the detected event is instantly
improve recognition accuracy. Our developed fire recognition sent to the owner’s smartphone, as shown in Fig. 1c(d). In the
model consisting of a number of dense blocks, where each DeepSafe system, MQTT is a lightweight, publish-subscribe
dense block consists of five layers and each layer has three network protocol over TCP/IP that transports sensing data
consecutive operations including Batch Normalization (BN), between onboard sensors and mobile devices.
Rectified Linear Unit (ReLU), and Convolution (Conv). The For desktop demonstration, cigarette lighters and fire pic-
transition layer is first adopted to reduce the number of feature tures can be used to simulate gas leakage and fire events in a
maps in the output of the last layer, and then the convolutional limited space, such as in a conference room, to demonstrate
and pooling layers are used for the final output of the dense our DeepSafe system, respectively. In addition, the abnormal
block. vibration level can be detected through the 3-axis accelerome-
In the control mode, DeepSafe can automatically block ter by flapping/shaking the IoT device. Furthermore, to provide
the gas source as detecting an abnormal event, where the remote monitoring, the real-time video captured by onboard
event-related warning message is simultaneously sent to the camera can be streamed to the owner’s smartphone. In par-
owner group. In addition, the kitchen owner can use his/her ticular, the owner can manually turn on/off the gas switch
smartphone to remotely watch the real-time streaming video controlled by the servo motor through his/her smartphone.
in the kitchen. Furthermore, even DeepSafe does not detect IV. ACKNOWLEDGMENTS
any abnormal event yet in the kitchen, the gas source can be
manually turned off through using a smartphone as necessary. This research is supported in part by MOST under Grant No.
106-2221-E-035-019-MY3 and 109-2221-E-035-062-MY3.
III. S YSTEM I MPLEMENTATION R EFERENCES
We have developed an Android-based DeepSafe prototype [1] L.-W. Chen, T.-P. Chen, H.-M. Chen, and M.-F. Tsai. Crowdsourced
consisting of a Raspberry Pi embedded system (as IoT device), Children Monitoring and Finding with Holding Up Detection Based on
Internet of Things Technologies. IEEE Sensors Journal, 19(24):12407-
NVIDIA Jetson Nano (as external GPU device), a note- 12417, Dec. 2019.
book computer (as data exchange server), and smartphones [2] J. Monteiro, R. Granada, R. C. Barros, and F. Meneguzzi. Deep Neural
(as owner platform). The Raspberry Pi embedded system is Networks for Kitchen Activity Recognition. International Joint Confer-
ence on Neural Networks (IJCNN), May 2017.
connected with the MQ2 gas sensor [7], ADXL345 3-axis [3] J.-M. Sun et al. Design and Implementation of Intelligent Kitchen System
accelerometer [8], and Logitech USB camera for abnormal Based on Internet of Things. International Conference on Smart Grid and
event detection. NVIDIA Jetson Nano is additionally attached Electrical Automation (ICSGEA), June 2018.
[4] S. Jamadagni, P. Sankpal, S. Patil, N. Chougule, and S. Gurav. Gas
to perform real-time fire intensity recognition in the IoT device Leakage and Fire Detection using Raspberry Pi. International Conference
(i.e., edge computing version). In particular, fire intensity can on Computing Methodologies and Communication (ICCMC), Mar. 2019.
be recognized in the DeepSafe server (i.e., cloud computing [5] S. Shrestha, V. P. K. Anne, and R. Chaitanya. IoT Based Smart Gas
Management System. International Conference on Trends in Electronics
version) if considering budget limitation without using any and Informatics (ICOEI), Apr. 2019.
externa GPU device. The servo motor SG90 [9] is used to [6] M. Ralevski and B. R. Stojkoska. IoT Based System for Detection
automatically block the gas source (as detecting an abnormal of Gas Leakage and House Fire in Smart Kitchen Environments. 27th
Telecommunications Forum (TELFOR), Nov. 2019.
event) and manually turn off the stove fire in a remote manner [7] Analog Gas Sensor (MQ2). http://www.energiazero.org/arduino sensori/
(as necessary). Analog Gas Sensor MQ2.pdf.
The IoT device of DeepSafe is installed on the gas stove [8] 3-Axis Digital Accelerometer (ADXL345). https://www.analog.com/
media/en/technical-documentation/data-sheets/ADXL345.pdf.
in a real kitchen environment, as shown in Fig. 1c(a), where [9] Servo Motor SG90. http://www.ee.ic.ac.uk/pcheung/teaching/DE1 EE/
the servo motor can control the gas switch that is connecting stores/sg90 datasheet.pdf.
the gas stove to gas pipeline, as shown in Fig. 1c(b). The [10] Message Queueing Telemetry Transport (MQTT). http://mqtt.org/

Authorized licensed use limited to: Chaitanya Bharathi Institute of Tech - HYDERABAD. Downloaded on February 09,2023 at 08:45:00 UTC from IEEE Xplore. Restrictions apply.

You might also like