Project
# | Title | Team Members | TA | Documents | Sponsor |
---|---|---|---|---|---|
33 | Supply and Demand Parking Meter |
Adam Barbato Nicholas Johanson |
Yuchen He TA | design_document0.pdf design_document0.pdf final_paper0.pdf presentation0.pdf proposal0.pdf |
|
Adam Barbato - barbato2 Nick Johanson - njohans2 Link to idea thread: https://courses.engr.illinois.edu/ece445/pace/view-topic.asp?id=13932 As the TAs no doubt know, we students seem to really enjoying trying to solve parking. Most students approach the problem with solutions to help find parking spots such as lights above available parking spots, or an app that shows available parking spots. However, we believe this fundamentally misunderstands the problem. Finding parking isn’t hard, it’s that in areas where people spend 30 minutes looking for a parking spot, the spot doesn’t exist, there isn’t enough parking; there’s too much demand for the supply of parking spots available. And, as any economics class will tell you, if there’s too much demand for the supply available, raise the price. As a solution, we propose a computerized parking meter that can dynamically change parking price as to target a certain percentage of parking spots remain open at all times. Our delivered parking meter will be constrained to only work on clear days, as to limit the scope of the project, and have three main components: the meter itself and the hardware and software on it, a software backend running on a remote server that uses data sent to it from the meter to determine proper prices using a PID-like control algorithm, and training data fabricated based on what data and research we can find about parking statistics. The meter itself will perform the functions of monitoring when and where cars are parked within its range, wirelessly transmitting that information to the backend server, and would be used to display that information to the user and process transactions, but we will not be implementing those features for this prototype. The hardware on the meter will consist of a prepackaged camera and wireless transmitter (Xbee, or like device) hooked up to a custom built microprocessing unit using, most likely, an ATMega processor on a custom PCB running arduino software. The camera will be statically positioned and will be used to determine if there are cars in each spot. This will be done by taking a picture every few minutes (to lower processing load) and running some rudimentary machine vision techniques, mainly taking an FFT, and comparing the results with a pre-taken picture of the area with no cars. The spots taken will then be transmitted to the backend. The backend will receive and monitor parking data it receives from the parking meters and adjust prices in an attempt to adjust the average parking density towards a set goal. For example, if the parking density for an area is too high prices may increase there and decrease somewhere nearby where the parking density is below the goal. We intend to have a PID-like control loop to control these prices with an underlying machine learning algorithm trained on the parking prices with the goal of better estimating parking densities in response to changes in pricing. The PID-like loop will primarily be proportionally driven so prices don’t change drastically if we are close to the parking density goal. The machine learning algorithm will be implemented as a neural network which over time will be trained on parking prices and attempt to estimate parking density given a distribution of prices. We hope to develop a system capable of estimating parking densities based on both time and day. Exact deliverables for the project will be a physical meter prototype that can detect cars in a stationary position, send car data to the server, and receive price information, a backend software platform for determining the best price for an area given training data or time to train itself based on the location, and a simulation of said backend software using fabricated training data based on our best guesses from the data available. |