Intelligent Trash Bin Management System –

initial design and implementation

Piotr Filarski1 , Piotr Niedziela1 , Maria Ganzha1,2 , and Marcin Paprzycki2

1
Warsaw University of Technology, Warsaw, Poland,
[email protected]
2
Systems Research Institute Polish Academy of Sciences, Warsaw, Poland
[email protected]

Abstract. One of interesting aspects of the Internet of Things (IoT)

is development of systems where sensors (and possibly actuators), allow
creation of inexpensive human-oriented solutions. In this context, we
consider “smart garbage collection”. Specifically, we have developed an
inexpensive prototype of a solution for poor communities (possibly Third
World countries). The proposed approach, consists of a trash bin proto-
type and an initial version of an application that allows optimisation of
trash pickup.

Keywords: Internet of Things, smart trash bin, garbage collection man-

agement

1 Introduction
As prices of sensors, actuators and computer hardware decrease, it is possible
to introduce inexpensive human-focused solutions. One of areas, where such
solutions are highly desirable, is garbage collection. Here, typically, garbage is
collected at certain days and times, e.g. every Monday, Wednesday and Friday,
before noon. This means that, each time the garbage truck is travelling the same
road, regardless if there is a reason to do so, or not (i.e. it will come to pick up
trash even if the trash bin is only half full and does not generate unpleasant
odours). As a result, fuel is consumed and unnecessary pollution generated. At
the same time, if the trash bin is full already on Wednesday in the evening (e.g.
because of a birthday party in the area), the truck will come only on Friday,
leading to trash piling up (with all unpleasant consequences of such situation).
The aim of this work is to present an initial prototype of an inexpensive smart
trash management system, consisting of a trash bin containing sensors and an
application that informs about the status of individual bins. To this effect we
proceed as follows. We start (in Section 2) with a brief description of the state
of the art. Next, in Section 3 we outline the proposed approach, following in
Section 3.1 with description of the smart bin prototype. Sections 3.2, 3.3 and 3.4
introduce the database and both the server and the web application support-
ing optimised trash collection. Finally, in Section 4 a simple approach to route
generation has been presented.
2 Piotr Filarski, et al.

2 State of the art

Let us start by briefly summarising the current state of the art in the area of
smart garbage collection. Here, we focus on (attempts at) actual deployments
rather than on research summarised in academic publications.
There are few companies, which provide waste management solutions. One
of them is Slovak Sensoneo 3 . They use ultrasonic sensors to measure trash fill
level (no weight measurement). Company declares that their solution reduces
cost of waste collection by at least 30% and carbon emission by up to 60%.
They also offer application, which informs about waste levels in all monitored
bins and enables finding the nearest available empty bin. Their planning solution
optimises waste collection routes based on data about fleets, depots and landfills.
The second company is Smartup Cities 4 . Their solution is based on wireless
ultrasonic sensors (installed in containers). Here, Google Maps API and data
from sensors allows generation of optimised collection routes. They also provide
a web and mobile platform to monitor waste level. The company claims that it
can save up to 50% of costs, by reducing operating hours and trucks maintenance.
Solution uses artificial intelligence and machine learning algorithms to optimise
collection routes and provide predictive analysis.
Next, the Ecube Labs 5 also offers solution based on ultrasonic sensors, capa-
ble of connecting to a cloud platform (named CCN). Sensors can communicate
via LoRaWAN or NB-IoT. Ecube Labs trash bins facilitate trash compacting,
claiming that this allows containers to hold up to 8 times more waste. Trash
bins can be powered by solar energy or AC power. The CCN provides moni-
toring environment, smart dashboard, analytics and a control centre. Additional
product named CCNx uses data analytics to optimize trash pickup routes and to
facilitate various monitoring functions. Their solutions are used, among others,
at the Dublin Airport, in Seoul and Washington D.C [6].
U.K. based Enevo 6 reduced costs of waste management at seven McDonalds
locations across Nottingham. Their technology automatically plans daily routes
for an entire collection fleet to make sure containers are collected as needed.
Finally, Ecobins 7 is a Polish company, which provides sensors for trash bins.
These sensors measure filling level, current position, tilt level and temperature
inside the trash bin.
All these solutions probably work in practice (our assessment is based only
on the promotional material found on their websites). However, using them on
a large-scale might be expensive. Additionally, presented solutions do not pro-
vide weight measurement and warning about fire in the trash bin. The latter is
particularly important as burning trash may result in large-scale fires.

3
https://sensoneo.com/
4
https://www.smartupcities.com/
5
https://www.ecubelabs.com/
6
https://www.enevo.com/
7
http://ecobins.pl/
 Intelligent Trash Bin Management System 3

3 Proposed approach

Our goal is to develop an inexpensive solution for waste management, based on

smart trash bins and a web-based application for trash monitoring and pickup
routing. Thus, we have identified the following key aspects of the sought solution:

– continuous monitoring of trash level in containers, i.e. to access detailed,

current and historical data of each container separately,
– data should consist of both the weight of the trash and the trash-fill level,
– generation of routes, including only these containers which require emptying,
– estimating, which containers will be filled in a given time-frame,
– sensing smoke to generate alerts (trash bin is on fire),
– Internet connectivity,= to communicate data to the central application,
– documenting pick up actions, recognised by the container.

Let us note that we have decided to use both the weight of trash, and the
level to which the container is filled. Trash weight provides extra information
that might be useful when trash pickup is planned. For instance, if a given bin
that should be filled with plastic is very heavy, a warning should be issued that
it is likely not to be filled according to the “requirements” / expectations.
Of course, these are basic requirements, based on a minimal number of sensors
that allow achieving the proposed functionality. As the price of sensors continues
to decrease, further sensing capabilities (e.g. establishing main components of
the waste, on the basis of detected chemical composition) that can be achieved
at the same price, should be considered. Finally, a GPS sensor might be mounted
to warn about potential theft (container movement).
Based on the above assumptions, we have developed an initial prototype of
a smart bin, consisting of (see, also, Figure 2):

– ultrasonic sensor detecting level of trash (cost of approximately 3 EUR)

– weight sensor (≈ 2.6 EUR)
– standard smoke detector (≈ 5.4 EUR)
– Raspberry Pi board with WiFi module (≈ 32 EUR)

Hence, the total cost of turning trash can into a smart bin was about 40
EUR. Recall that this is a preliminary solution. Prototype might use a micro-
controller with a mounted LoRa module, thus reducing the total cost to ≈ 17
EUR. Furthermore, in case of mass production, price of a single set of electronics
would go down considerably (above prices are based on what we have actually
paid for individual components purchased from online retailers).
For the used sensors accuracy of distance measurement from containers cover
to the filling level is ≈ 1 centimetre, while the weight data has accuracy of ≈ 10
grams.
Obviously, the question arises: how the proposed trash bin could be pow-
ered? Due to the high demand for electricity, the prototype is currently supplied
with “stationary power”. However, after replacing Raspberry Pi with a micro-
controller, the prototype might become battery-ready. In this case, it will also
4 Piotr Filarski, et al.

send information (to the central management unit) about its battery level. It
might be possible to use a solar panel, but it would considerably raise the total
cost. Nevertheless, in countries with a lot of sunshine, this may be a reasonable
solution (to be evaluated from the financial perspective).
The developed prototype is similar in appearance to a regular municipal
waste container (Figure 2). It consists of a basket and a cover. It has been
designed so that electronics are not visible to passers-by. The prototype main-
tenance is supported by its design, which enables easy access to the electronic
components, if repairing or replacing them is needed. The bin sends data, as
long as it is connected to a power supply and has access to the Internet, via
a WiFi network. Instead of WiFi, prototype might use LoRaWAN to reduce
power consumption [8]. In the case of radio connectivity a field gateway could
be developed, with wired Internet connection and a range of few kilometres.
Separately, to reduce energy consumption, the container sends data about its
state only when necessary (when new waste is detected, collection event occurred,
or fire has been detected). This is a standard approach for the targeted battery-
power-based operation.
To support the garbage collection service (end-user), the developed solution
includes a web application. Based on data generated by trash bins, it allows
more effective trash collecting. Specifically, it allows user to view containers on
the map and access their detailed data (current filling level and weight of waste).
Moreover, each container records recognised emptying actions, available within
the application. Finally, a very preliminary mechanism of selection of optimised
routes for garbage collection has been developed.
In order to verify operation of the application, a separate simulator module
has been created. It is a console application, designed to simulate filling of trash
bins located in Warsaw.
Figure 1 depicts the system architecture. It is composed of the above men-
tioned components: Web Application, smart waste Bin, Simulator, Server and
a Database. The Application communicates with the Server in order to display
information generated by the Bin. The Simulator emulates existence and be-
haviour of multiple smart bins.

3.1 The prototype of the smart waste bin

Let us now describe, in more detail, the hardware of the prototype of the smart
bin. The core element is a Raspberry Pi 3B+, a single board computer that runs
Raspbian, which is a free operating system based on Debian and optimised for
the Raspberry Pi hardware [9]. To the board we have connected three sensors:
load cell (NA27), distance sensor (HC-SR04) and smoke detector (MQ-2).
To provide basic information about the state of the bin, two LED diodes have
been used. The circuits were built on a bread board, which is a construction
base for wiring and and prototyping of electronics. The prototype bin is shown
in Figure 2.
Let us now describe, each of these components. The MQ-2 smoke sensor
([10])requires connection via a TTL (Transistor-Transistor Logic) converter to
 Intelligent Trash Bin Management System 5

Fig. 1. System architecture diagram.

Fig. 2. Prototype of the smart waste bin.

6 Piotr Filarski, et al.

reduce the signal voltage to acceptable for a GPIO Rapsberry Pi interface. The
5V signal is reduced to 3.3V. Then, an analog signal is converted to a digital one,
using an ADC converter (MCP3008). Finally, the sensor data is read via GPIO
Raspberry Pi input. The sensor data signal is a voltage signal that allows to de-
termine concentration of gases, and thus the smoke. Here, as a gas concentration
in the air increases, the resistance of the internal sensors components increases
as well, signalling smoke. The HC-SR04 ultrasonic distance sensor ([11]) also
requires a logic levels converter to reduce the data signal voltage to the one
acceptable by the GPIO interface. This sensor provides a TRIG input and an
ECHO output. Setting high state on TRIG input, for 10s, triggers the distance
measurement. The output is a signal, duration of which is proportional to the
measured distance.
The load cell requires connection via the HX711 operational amplifier ([12]),
which is sensitive to changes in the load cells resistance. Load cell bends under
applied force, changing the resistance. For communication with Raspberry Pi it
uses two lines: SCLK (clock) for synchronisation and DATA (data) with digital
weight values.
The described connections and the single board computer have been mounted
under the containers cover, as presented in Figure 3. Two LED diodes visible on
the bread board, enable reading the operating state of the prototype, without
need of external monitor connection.

Fig. 3. Electronic device used in the prototype

The distance and smoke sensors are attached to the other side of the metal
grille, providing correct readings from the containers interior and, at the same
time, limiting the length of wires. The sensors are presented in Figure 4.
 Intelligent Trash Bin Management System 7

Fig. 4. Sensors; from the left: smoke sensor, distance sensor

A prototypes software, which is reading data from sensors and sending it

to the Server was implemented in Python 3 programming language. The script
calibrates sensors and checks values returned by them in the loop (which is ex-
ecuted every 8 seconds; parameter that can be adapted as needed). When a
significant change is detected (what constitutes a significant change is a param-
eter adjustable for each sensor separately). New data is sent to the Server using
HTTP protocol and JSON format messages.
The script has also an additional sleeping time of 2 seconds after new data
detection. Then data is read again, to eliminate situation when falling garbage
“deceives” the sensor reading. The reading is treated as correct, if both reads are
the same. In this case, data is sent to the Server. If the second read is different,
then data will be read again after 2 seconds. In the case when 5 consecutive
reads (every 2 seconds) generate different values, warning is sent to the Server
and the reading process is reset to the basic mode (repetition every 8 seconds).
Obviously, in all cases, time between reads is a parameter that can be tuned on
the basis of real-world observations.
The prototype starts to work as soon as it is connected to a 5V power supply.
Single blinking of the red and blue LEDs indicates automatic script execution.
Next, when the sensors calibration completes, user can observe the blue diode
blinking. This indicates successive iterations of the main loop and readings from
sensors. When new garbage is detected, the red diode blinks once. However,
when garbage collection or fire are detected, the red diode blinks twice.

3.2 Database
In the system, it is necessary to store data needed by the Application. For this
purpose, a relational database was created. Here, note that while the Database
is described as it was placed within the Server (which is the way it was initially
implemented), a cloud-based solution is also possible. As a matter of fact, it may
be more desirable, once a real-life large-scale deployment is to be instantiated.
However, this is only a technical issue that does not influence the design of the
proposed system. The Database scheme is presented in Figure 5.
8 Piotr Filarski, et al.

Fig. 5. Database diagram

The Bins table stores static data, describing states of individual trash bins.
The BinId column is auto-incremented, unique key of each waste bin. The Lon
and Lat (longitude and latitude) specify the exact location, where the trash
bin can be found. The MaxFillLevel contains information about the maximum
capacity of each garbage bin (height in centimetres). The City is the city where
the container is located, while Address is the street name along with the building
number nearest to the bin and RecentDataId is the foreign key of the most
current data telemetry from the DataFromBins table.
The DataFromBins table stores the telemetry data of a given container. The
Id is a unique, self-incremented key, the Distance is the distance from the bins
cover to the level of garbage (in centimetres), the BinId is the foreign key of the
garbage bin, the Weight is the weight of the garbage (in grams) and the Date is
the time of receiving given data, in the standard format.
The Events table stores information about anomalies and events that have
occurred, such as a garbage fire. Here, the Id is a unique, self-incremented key,
the EventName is the name of an event (any text), the BinId is the foreign key
of a bin, which this event has affected and Date is a time of receiving this data
in a standard format.
The GarbageStations table stores information about the garbage trucks sta-
tions. The Id is unique, auto-incremented record key. Lon and Lat are the exact
location, in which given station is located. City is the city where the station
is located and Address is the name of the street with the number of a stations
building as a string type.
The GarbageTrucks table stores garbage trucks information. The Id is a
unique, auto-incremented key. The GarbageStationId is a foreign key pointing
to the station, to which the given garbage truck belongs. Capacity is the capacity
of the garbage truck, defined as the maximum number of waste bins, which can
be emptied by a given truck.
 Intelligent Trash Bin Management System 9

Data generated, and used by the system can be divided into two types: static
data (information about garbage containers, e.g. Location, information about
garbage trucks, e.g. Capacity) and time series data, i.e. telemetry data generated
by sensors mounted in containers.

3.3 Server

An application is running on a remote machine, hereinafter referred to as a

Server, which is used for communication between smart waste bins and the web
application. Server should be considered as a central computing resource and
might be replaced with a cloud solution.
The Server performs reading and writing of data from and to the Database.
It also performs data processing such as computing percentage filling level from
the distance values and thus provides the logic of the Application.
The Server has been created according to the MVC pattern (Model-View-
Controller) which is an architectural pattern used to organise the structure of
applications with graphical user interfaces. MVC divides applications into three
main components:

– Model – is a certain representation of the problem or application logic.

– View – describes how to display a certain part of the model within the
interface.
– Controller – accepts input data from the user and responds to her/his actions
by managing model updates and refreshing views.

The Server is based on the REST concept (Representational State Transfer).

REST is a set of practices, which determine how network services should be
implemented and is based on entities. Operations on entities are performed using
HTTP queries. Entities can be modified using different query types. Types of
queries are defined by verbs: Get, Post, Put and Delete. These verbs together
with the URL address define what operations will be performed on a given entity.
The Server has been implemented in ASP.NET MVC technology using C#
language. The application is divided on the following layers: presentation, busi-
ness logic and database access. Presentation layer provides Server communica-
tion with clients (Bins and web Application). Each controller is assigned to one
table in the Database. In addition, there is a controller responsible for display-
ing the main view. Business logic layer processes the data and prepares it for
sending. In this layer each class corresponds to one of the Database tables and
provides data processing on it. In particular, smart bin data is prepared to be
displayed on the map. The Database access layer is responsible for updating
data in the Database and returning the current data. Access to the Database
is provided thanks to the EntityFramework. The Server also sends information
about received data, so that user can observe changes (as they happen). This is
ensured with the SignalR library. ISBContext interface is used to communicate
with the Database.
10 Piotr Filarski, et al.

3.4 The Web Application

The aim of the Application is to present data processed by the Server. Applica-
tion retrieves data using Ajax queries. Smart Bins are displayed on the map as
marker points in two different colours, depending on whether the Bin is full or
not. When clicking on a given marker, in the right panel, detailed information
about the selected Bin is displayed. The most recent weight and fill-level data
are presented there as well. Moreover, user can view a table containing events
that have occurred (e.g. garbage arson, emptying event). Thanks to the Sig-
nalR technology, data in the application is being updated without reloading the
page [7]. The corresponding module receives data from the Server and changes
data in JavaScript. In addition, the application displays the routes of collection
trips using Google Maps API. We decided to use Google Maps API, because it is
the market leader. Nevertheless, it should be stressed that, thanks to the mod-
ular design of our application, Google Maps can be easily replace with another
mapping software, e.g. Open Street Maps.
A map with generated smart Bins markers is shown in Figure 6. Blue markers
correspond to the bins that are not completely filled up, whereas orange markers
represent the bins, which are already full. There is an information panel on the
right side of the application, which appears when one of the containers on the
map is selected. It contains data about selected Bin. At the top of the panel,
basic information is displayed, such as identification number, filling level, or
weight of waste. Below, there is a location of a selected Bin and the date of
the last message received form it. At the bottom of the panel, there is a table
containing the last 15 filling level updates received. Below, there is another list
of the last 15 events received for the selected Bin along with the date and time
of the event.

Fig. 6. A map with generated smart waste bins markers

Garbage collection routes are shown in Figure 7. This is the view of the
application, when the Show stations option has been chosen from the menu,
 Intelligent Trash Bin Management System 11

followed by clicking View routes in the right panel. The information on how
many trucks are needed is also displayed. On the map, except routes, there are
also full Bins presented, which are to be visited during the service trips with
starting points as trucks stations (purple markers).

Fig. 7. Garbage collection routes

In Figure 8 a list of Bins, which are likely to fill up within the next two hours is
displayed. This application view appears while choosing “Show analysis” option
from the menu. On the map, yellow markers correspond to the bins shown in
table in the right panel. The prediction is based on the history data and has been
modelled as the average filling speed of the bins. The average was calculated as
total filling level of trash collected over a period of time divided by this time.
Obviously, this is a very simplistic approach and in a real-life deployment it
would have to be replaced by a more sophisticated one, e.g. based on machine
learning. However, since our data is only simulated, using a simple averaging is
justified, as a way of illustrating how the proposed approach might work.
To ensure proper usability, the Application is adapted to the standard screen
sizes of PC stations and portable computers. The Application has an intuitive
and easy-to-use dashboard and works on Mozilla Firefox (from version 57.0) and
Google Chrome (from version 62.0). It should work also on other browsers, but
they have not been thoroughly tested.

3.5 Simulator
Due to labour-intesive process of building physical prototypes of smart bins,
simulation environment (Simulator ) has been created. Simulator of smart bins
in Warsaw is a console application implemented in C# language and requires
a .NET platform as a runtime environment. In order to retrieve current data
about simulated containers the Simulator requires a Database connection. The
supported input arguments of the program are decimal numbers from 1 up to
12 Piotr Filarski, et al.

Fig. 8. Bins, which are probable to fill up within two hours

10, as a simulation speed, and parameter c for simulation of cleaning service

collection trips. A parameter with a value of 1 means generating about one
action per second, while parameter 10 about 6 actions per second. In the case of
a numeric parameter, the application simulates throwing garbage into bins by
passers-by in a loop. Pseudo-random garbage data is sent, representing readings
from three sensors: distance from containers cover to a new level of filling, current
weight, and messages about random events (garbage fire). Iterations of simulated
actions are executed in intervals, timing of which is controlled by the input
parameter. In addition to generating data of smart waste bins, the application
also simulates the real-time flow. It allows continuous simulation even if simulator
is closed and restarted. If simulator is run with the c parameter, the collection
of garbage is simulated by sending event messages of emptying bins. From the
Server ’s point of view the Simulator cannot be distinguished from the physical
waste container prototype.

4 Generating services collection routes

Let us now illustrate how the proposed approach can help generating trash
bin collection routes. Here, let us emphazise that we are aware of a number of
works that deal with optimal route generation, including optimal truck dispatch
routing (see, for instance, [1–5]). However, finding the best algorithm for route
generation was out of the scope of our current contribution. Nevertheless, we can
already illustrate that availability of a smart trash bin, like the one described
above, can reduce costs of waste collection.
A simple way to achieve this goal is to take into account only these trash
bins, which filling levels exceeded a set threshold. This solution eliminates the
necessity of visiting and emptying unfilled waste bins. We assume that each city
has several garbage stations, in which trucks are stationed. However, since each
trucks station has a limited number of trucks available, it is not sufficient to use
 Intelligent Trash Bin Management System 13

a trivial solution, in which the bin is served by the nearest station. Let us look
into some more details of this issue.

4.1 Routes calculation scheme

To illustrate potential positive effect of the proposed approach, let us assume

that only bins which are at least 80% full are considered for pick-up. Assigning
containers to the garbage truck stations takes into account limited number of
available trucks, as well as their capacities. In this way, bins are automatically
divided into collection areas.
The proposed approach, starts from assigning bins, for which assignment to
another station (not the nearest one) would have the most significant negative
effect (largest distance difference). In other words, bins that are in the closest
proximity of each station are allocated first. As the assignment process pro-
gresses, bins that are located “in between” stations are considered (difference in
distance to the possible pickup stations is decreasing with each assigned bin).
Moreover, starting form a certain moment, the capacity limitation comes to play,
as stations may not be able to serve more bins. However, this means that these
bins that are “left” are also “close” to the other pickup stations. After all bins are
assigned, the routes are displayed in different colours and the DirectionsRenderer
class from Google Maps API has been used. Functionality has been obtained by
creating a list of full containers in JavaScript and sending a query to Google
Maps API.
Figure 9 visualises the division of filled containers into collection areas. Three
truck stations were instantiated for this simulation. The area corresponding the
first station contains 24 bins, the next areas contains 10 and 6 bins. While
the number of bins in each area differs, they are the closest to their respective
collection stations.

Fig. 9. Visualizations of the division of filled containers into 3 areas

14 Piotr Filarski, et al.

In the case of two truck stations, Figure 10 presents the new division. In this
example the first station is to serve 29 bins and the second one 11. This shows
how the proposed algorithm works for different number of stations.

Fig. 10. Visualizations of the division of filled containers into 2 areas

Simulations with varying numbers of smart bins and truck stations (with
varying total capacities of available trucks) have been performed and correctness
of the developed algorithm verified.

5 Concluding remarks

The aim of our work was to show the possibilities and benefits brought by analy-
sis of data collected from sensors placed in urban waste containers. The goal was
achieved by creating a solution consisting of three elements. The first of them was
an inexpensive smart bin prototype, which collects data about filling level, waste
weight and is detecting potential arson. Implemented simulator enables testing
the solution at the city-scale, while the web application presents opportunities
of data analysis. In order for this solution to be ready for go-to-market process
some improvements and extensions have to be considered. A more advanced al-
gorithm of route calculation would lead to better optimisation of costs incurred
by the city. It should be beneficial to extend the current division algorithm by
adding (in the second stage) optimisation of routing in each of the areas that
have been established in the area division stage (following results found in perti-
nent literature). Moreover, prototype requires further development of electronic
components, for which miniaturisation would be necessary. Smart bin should be
powered by a microcontroller and the circuits should be created based on a ded-
icated PCB board. In addition, communication technology should be updated
so that a greater range and less power consumption could be achieved. Taking
it all into consideration the second version of the prototype would be based on
 Intelligent Trash Bin Management System 15

LoRa32U4 II board, bringing miniaturisation, battery supply and radio connec-

tivity. It seems beneficial to use artificial intelligence algorithms for prediction of
filling as well to correlate data from sensors with external data, such as people
concentration in the city.

References
1. G.B.Dantzig, J.H.Ramser, The Truck Dispatching Problem, Management Science,
(1959)
2. M.L. Optimal Solution of Vehicle Routing Problems Using Minimum K-trees, Op-
erations Research, (1994)
3. J.K. Lenstra, R.Kan, Complexity of vehicle routing and scheduling problems, Net-
work, (1981)
4. T.K.Ralphs, L.Kopman, W.R. Pulleyblank, L.E.Trotter, On the capacitated vehicle
routing problem, Mathematical programming, (2003)
5. E. Uchoa, D.Pecin. A.Pessoa. M.Poggi etc, New benchmark Instances for the Capac-
itated Vehicle Routing Problem, European Journal of Operational Research, (2016)
6. Ecube Labs Case Studies https://www.ecubelabs.com/references/. Last accessed 6
Apr 2019
7. Microsoft ASP.NET SignalR Homepage https://dotnet.microsoft.com/apps/
aspnet/real-time. Last accessed 6 Apr 2019
8. S. L. Martinet, 3 Differences between LoRa and WiFi. Medium (2018), medium.
com/@samLmartinet/lora-vs-wifi-3-questions-d9c93137fca. Accessed 6 Apr 2019
9. Raspbian Homepage, https://www.raspbian.org/. Last accessed 6 Apr 2019
10. Learning about Electronics, http://www.learningaboutelectronics.com/Articles/
MQ-2-smoke-sensor-circuit-with-raspberry-pi.php. Last accessed 6 Apr 2019
11. Codelectron, http://codelectron.com/measure-distance-ultrasonic-sensor-pi-hc-sr04/.
Last accessed 6 Apr 2019
12. Rasberry Pi Tutorials, https://tutorials-raspberrypi.com/
digital-raspberry-pi-scale-weight-sensor-hx711/. Last accessed 6 Apr 2019