Você está na página 1de 6

Austen Roberson

Mrs. Curtin

Intern/Mentor G/T

24 April 2019

The Cure for Parking Pain

Parking pain is an issue that is prevalent yet rarely discussed in today’s society. It is an

issue that can lead to missed appointments, lost wages, and increased stress (Oldfield, 2017). An

intelligent program can help to mitigate these negative effects by facilitating the ways that people

can access parking. To solve the problem of lack of parking, a computer vision machine learning

software, utilizing optical passive sensors can be created.

The lack of parking is a issue that can affect all people in a negative fashion. Even the

simple action of searching for a parking space can cost drivers money. In addition to the inflated

parking costs, drivers also have to suffer lost wages and fuel trying to find parking in crowded

cities and lots. While attempting to find a parking space, drivers spend $345 in wasted time, fuel

and emissions per year. These costs take into account missed wages and the average American

fuel prices (Oldfield, 2017). The economic impacts of parking are often overlooked. With a more

efficient parking solution, drivers save money in gas prices and miss less work time.

In addition to these monetary detriments, many drivers report that a lack of parking

causes negative stress and emotion. Nearly two-thirds of drivers surveyed feel stressed searching

for parking. This stress is induced by a variety of factors, including missed appointments as

“nearly half (42%) [of drivers surveyed] missed an appointment [due to parking problems].”

Lack of parking has also been identified as a trigger for road rage for one-quarter of the drivers

(Oldfield, 2017). This has led to more than half of drivers avoiding driving to a location due to
struggles of finding parking. According to a study by the parking analysis firm INRIX, “39% of

respondents avoided shopping destinations because of the lack of parking, 27% didn’t drive to

airports, 26% skipped leisure/sports activities and 21% avoided commuting to work. A surprising

20% of American motorists surveyed did not drive to the doctor’s office or hospital due to

parking issues.” The mental effects of parking are equally as important as the economic impact.

The fact that stress discourages drivers from travelling results in fewer consumers available.

Road rage has also been cited as a catalyst for car accidents and aggressive driving. If parking

became easier to find, motorists would be able to reduce their stress. They could also be

encouraged to travel to previously avoided destinations.

The negative effects of the parking search are not only limited to drivers. Driving to find

parking also has an adverse effect on the environment. According to studies performed

nationwide, most of the driving in cities is dedicated to finding a parking space. Anywhere

between 30 to 60 percent of the cars driving downtown city centers are cruising to find an open

space to claim. This action results in 47,000 gallons of gas wasted per year and 730 tons of

carbon dioxide generated (Thompson, 2016). Lack of parking affects many facets of life. This

unnecessary action of cruising for parking is causing air pollution and the corruption of the

environment. With a new solution for parking pains, the process can become more efficient and

reduce harmful environmental effects.

This new solution has to be powered by new developments in computer software - such

as computer vision. Computer vision programs have been proven to be able to solve many real

world image recognition problems. One example is the lane detection for driving assistance and

self driving car technologies. Using the fuzzy method in computer vision, the integration of
scanning and image processing algorithms (SIP) can accurately and efficiently locate and define

lanes. When using this method, The accuracy of the lane detection is 95% and the average time

taken to detect the lanes in each frame is 16.7639  ms (Zhang et al., 2010). Not only have these

programs proven to be accurate, they take less than 1/100th of a second to respond with new data

resulting in nearly real time updates for the driver. This accuracy and speed will be crucial in the

effort to find parking. It is imperative that drivers are updated as quickly as possible to reduce the

need to cruise around the lot. Accuracy is necessary to increase the users trust in the system. If

drivers are led to a spot that is not open, they will begin to doubt the capabilities of the program.

Pure computer vision is not enough to detect spaces in a parking lot. These programs

often employ the Hough Transform as a reliable method to detect both straight and uneven lines

in various image frame. The Hough Transform generates parameter values m and b in the line

equation y = mx + b for all lines that may pass through any detected point in an image. It then

draws all possible lines through every point then votes for its most common m and b values.

These votes are then accumulated and scanned to see which values of m and b parameters got the

most votes (Dawson, 2014). The parameters with the most votes give the equation of a line in the

image. In parking lots, this technology can be utilized to differentiate the spaces that distinguish

individual parking spaces. By sensing the parallel lines, a parking space can be identified as a

rectangular segment in an image. Once the segments are defined, the parking lot can be mapped

based on each space.

Computer vision systems can be used to recognize and classify many other objects in an

image as well. Many systems can receive data from images and identify distant objects better

than humans can. One of the most popular applications of computer vision is in facial
recognition, where it has been tirelessly implemented and improved. The accuracy of Facebook’s

facial recognition software is around 89%–94% when viewing faces from straight on (Savage,

2016). Computer vision software and techniques are implemented in advanced autonomous

vehicles, robots, drones, smartphones and more. The increase in accuracy has proven that

computer vision solutions are ready for real world implementation. Creating a parking solution

with this type of program ensures reliability.

Computer vision systems are steadily improving and may soon outpace human abilities.

These systems used to have error rate of approximately 26% five years ago; today, the error is

down to 5%–6% (Savage, 2016). A large part of what has made such rapid success possible is

the continuous development of training data sets with hundreds of thousands of images that are

hand-labeled by almost 50,000 people. Due to their vast access to such information, experts

predict that they should be able to always identify a breed of dog or the make and model year of

a car within a few years (Savage, 2016). With the continual improvement of the computer vision

software, in a few years computers may be more reliable than humans when attempting to

process images and/or video instantaneously.

Machine learning is able to be seamlessly integrated into most computer vision programs

and improve their capabilities. Machine learning allows a program to be able to learn and self

regulate. A neural network can learn what an object looks like and how to differentiate it from

others based on data and images (Wilson, 2015). Over time, certain neurons in the deep learning

network become better than others at recognizing and generating specific things (Greengard,

2017). Like the human brain, a neural network learns with the more images that it is exposed to,
over time it begins to generate its own parameters to more accurately define what an object is

and what it is not.

The largest draw of machine learning is that it allows the system to be able to solve

non-trivial problems. Such problems are usually characterized by several factors: the data are

multidimensional; the underlying pattern is complex (Schölkopf, 2015). Problems may be

changeable and/or nonlinear in that the definition of the problem may evolve over time. A

mechanistic understanding of the problem may be nonexistent and the human programmer is

unable to define exactly what the program will always be looking for. The goal of the learning

system is to determine which action will maximize an accumulated future reward. (Schölkopf,

2015). In a method called Q-learning, Q represents the accumulated future reward and the

computer will choose its actions based on the predicted value of Q for each action (Schölkopf,

2015). The reinforcement from a human operator can also play a significant role when training

these supervised programs to make sure the computer is retaining the correct data. This ensures

that the program is always improving and becoming continually more accurate.

Computer vision will be used to immediately address the problem, and machine learning

can be added to continuously perfect this solution. Ultimately when combined, these two

programming methods will create a robust solution that can be adapted to fit in parking lots

around the world. With a successful program, nobody will need to miss another appointment and

the cure for parking pain can be found.


Works Cited

Dawson, B. (2014). The Hough Transform in Machine Vision. ​Quality,​ ​53(​ 11), 26–30.

Greengard, S. (2017). It’s All About Image. ​Communications of the ACM,​ ​60(​ 9), 13–15

Jindong Zhang, Xiaoyan Jia, & Jinfeng Li. (2015). Integration of scanning and image processing

algorithms for lane detection based on fuzzy method. ​Journal of Intelligent & Fuzzy

Systems,​ ​29(​ 6), 2779–2786.

Malik, J. (2017). What Led Computer Vision to Deep Learning? ​Communications of the ACM,​

60​(6), 82–83.

Oldfield, P. (2017, July 11). Searching for Parking Costs Americans $73 Billion a Year.

Savage, N. (2016). Seeing More Clearly. ​Communications of the ACM,​ ​59(​ 1), 20–22.

Schölkopf, B. (2015). Artificial intelligence: Learning to see and act. ​Nature​, ​518​(7540), 486.

Thompson, C. (2016). No Parking Here. ​Mother Jones,​ ​41(​ 1), 16.

Wilson, A. (2015). Machine learning leverages image classification techniques. ​Vision Systems

Design​, ​20​(2), 31–33.

Você também pode gostar