This document provides an overview of the Smart Parking project at Intel-NTU Connected Context Computing Center. The project aims to create low-power self-reliant parking sensors that can detect if a parking spot is occupied for 3-4 years with minimal human interaction. The sensors use temperature and magnetic field readings with machine learning to classify occupancy. The project has faced challenges with unreliable hardware but proposes solutions like revising boards and establishing wireless communication between sensors and servers. The author contributes to hardware development and testing, troubleshooting faulty boards, and learning about wireless communication protocols.

1. Intel-NTU
Connected Context Computing Center:
Smart Parking
Written By: Josh Shih
Supervisor: Chuang-Wen You
Mentors: Allen Wu and Ivan Huang
Date: August 10, 2015

2. Introduction
To redesign a current technology to become more self-reliant and most
importantly, low-power. Smart Parking is a lab group that is currently working on
a part of the Internet of Things (IoT); as the name suggests, the group focuses on
the parking of regular-sized cars. The manufactured product is still undecided, but
for prototyping Smart Parking has gone forward and collected data using with their
own nodes that contain sensors within an industrial-standard waterproof box. This
node is meant to be at the surface of the ground (under supposed vehicles),
wireless, and is made to be situated in one place for a prolonged period of time
(3~4 years) with minimal human interaction.
Problem Statement/Research Challenges
The question is simple, “Is the car in this parking spot? Or is it not?” To answer,
Smart Parking has three sensors: one on-board temperature sensor, one off-board
temperature sensor, and lastly, one magnetic field sensor. All sensors write their
own values onto an SD-card every 10 seconds along with timestamp.
Smart Parking’s method of answering the previous question is with Support Vector
Machine (SVM). This method mainly takes advantage of Machine Learning; it’s
exactly as it sounds, as time goes on and more data is collected the machine can
then distinguish between multiple classes (types). Since the number of answers to
the question is two, only two classes are needed (“yes” and “no”). To properly use
SVM, the predicted values must be well-defined and that can only be done with
proper data values (sensor values) and a large enough sample size. Although this
seems relatively simple in context, actual implementation is quite a feat and took
up a good amount of time to properly implement.
One of the earlier hypothesis theorized that the temperatures would rise while the
car was directly overhead, but was quickly thwarted when results proved the
assumption to be false. Albeit their theory was wrong, a greater use was found in
using the temperature sensors in detecting the presence of a vehicle. Temperatures
dropping rapidly due to the nodes being covered by the car proved to be quite
telling and useful.
There are many challenges this group has faced and mostly have lied on the
hardware side as opposed to the rigorously tested code in the software spectrum.

3. Firstly, the boards that are currently in use are custom-crafted and thus one-of-a-
kind sensor boards. Once these boards have been exhausted, the next batch won’t
come in for a while since revisions also need to be made to make the product more
robust. Next big issue would have to be the MSP430 USB-Debug-Interface; the
main tool that enables users to manipulate and code the powerful microcontroller
that is the MSP430. The tool has proven to be quite flawed and the group found
themselves struggling to pinpoint direct causes of failure since the debug-interface
wasn’t always reliable. Since early July 2015, many debug-interfaces have been
labeled as flawed and are not sure why. Next is the indoor testing of our boards and
their ability to collect data. After a couple attempts to collect outdoor data, or “live
samples”, Smart Parking found themselves reading corrupted data from faulty
boards that weren’t rigorously tested indoor before deploying them out on a two-
day excursion. The loss of data from one flawed board is not detrimental to the
project, but as more and more boards appear to be under trouble the cumulative
losses are proving to greatly hinder progress. Lastly would be the wireless
capability of our nodes. Currently (8/10/2015), our boards cannot communicate
with a server or with each other. This problem in itself stunts the idea of IoT
completely.
To this day, what causes some boards to work and others not has yet to be figured.
This is a cause for concern because without certainty in our work and products
there is no way to move forward.
Proposed Solutions
The answers to most of these hardware obstacles are quite clear. The newer boards
that are to arrive should be heavily revised and made to optimize efficiency. Isolate
the “jump” issue with some boards where the node can’t initialize without the
touching of the oscillator. Figure if the debug-interface tools are flawed within
themselves or incompatible with the machines we’re connecting them with. Indoor
testing involving battery-powered nodes to check if the values are being read and
written correctly onto the SD-card.
Many of these solutions are underway and are sure to overcome the obstacles
before them (just not within the summer’s timeframe). Most importantly, wireless
communication between server and nodes is getting closer and closer to

4. completion. With this hurdle passed, the group can easily identify problems before
retrieving their boxes every two days.
Contributions to the Project
My contributions mainly include hardware. The soldering of the microcontroller,
sensors, capacitors, resistors and etc., on the boards. Testing the connections on all
the pins of the MSP430 and JTAG. Anything that is to be connected to the board at
deployment has at least been surveilled once by myself. It’s come to the point
where I know the values and locations of each capacitors, because I’ve made at
least five nodes and three boxes. Present day we have four parking spots being
tested and I am responsible for the creation of back-ups in case previous boards run
into trouble. Since we’re in prototyping, the boxes are picked up and deployed in
two or three day intervals; I also helped in doing this, allowing others to pursue
other ends of this project. Currently three of the four active boxes were made by
myself. I am also doing most of the troubleshooting in faulty boards, but if I am
stuck my mentors and advisor are always more than happy to help.
I have also been learning Zigbee, our means of wireless communication, to further
understand how to receive and read packets of information between themselves.
These Zigbee’s are key to Smart Parking’s success as an IoT group. Only recently
was I introduced to this powerful tool, so I have been reading up code and helping
Ivan spot the crucial parts we need to ensure a strong connection.
Aside from the hardware with my mentor Allen, I have also been a part of
language development with Ivan. Learning from both sides have given me great
insight on my career path. Although I’m not an engineer by any means, I found
myself enjoying the time I have in my workstation with a soldering iron.
The language the MSP430 reads is typically C due to the low memory limits of the
MCU. We have tried implementing C++, but I advised against since C++ hierarchy
does get quite complicated and could devastate the functionality of the MSP430.
Conclusion
The methods in gathering information and quality of the hardware itself already
make the prototype a contender with other parking devices known to date. There

5. are no self-reliant parking devices that can communicate with a server for longer
than 3~4 years; this is what makes this group on the cusp of something great.
Smart Parking has a clear idea of what they would ultimately market and have an
unwavering vision of their product becoming a game-changer in the world of IoT.
Their dedication as a three person group is inspiring to say the least. With more
data coming in and more reliable hardware on its way, the parking devices Smart
Parking develop will only establish NTU further as a contender for one of Asia’s
top universities.