License-Recognition-Based Parking Toll System

By: Felix Kelly-Yuoh, Samuel Pena, 

Jacob Rabinowitz, & Christopher Tamay-Chauca

Instructor: Susan Delamare

December 13, 2023

The City College of New York

Summary

Technology keeps on advancing, and as it does, older tech should be replaced. Parking meters have been around since the 1930s and are responsible for easing traffic congestion and providing infrastructure funding, but they are getting too old and ineffective for the modern era. This proposal suggests a new and revolutionary idea for parking garages. The Parkcam system is designed to take the license plates of vehicles and then charge the owner directly. It will be designed to compare the vehicles’ entry and exit times through multiple cameras and charge them appropriately. This system will benefit everyday drivers by having fast parking and no risk of being towed. In fact, such a system already exists in Taipei, Taiwan, using license recognition to charge their drivers for parking, which establishes a positive precedent. With all the estimations possible, the costs for this project are between $25,516 and $46,196 but should make up for its costs by reducing human error in parking tolls.

Authors’ Note

This paper was prepared for English 21007, taught by Professor Susan Delamare.

Table of Contents

1. Introduction ……………………………………………………………………………………………………………….. 3

2. Objectives ………………………………………………………………………………………………………………….. 5

3. Preliminary Literature Review ……………………………………………………………………………………… 6

4. Technical Description of Process ………………………………………………………………………………… 10

5. Budget …………………………………………………………………………………………………………………….. 13

6. References ……………………………………………………………………………………………………………….. 15

7. Appendix A – Task Schedule Chart ……………………………………………………………………………… 18

8. Appendix B – Camera Details …………………………………………………………………………………….. 19

List of Figures

Fig. 1. General Process Flow Chart …………………………………………………………………………………. 12

Fig. 2. Project Timeline …………………………………………………………………………………………………. 18

Fig. 3. Camera Dimensions ……………………………………………………………………………………………. 19

Fig. 4. Camera ……………………………………………………………………………………………………………… 19

List of Tables

Table 1. Project Budget Breakdown ……………………………………………………………………………….. 13

Introduction 

Parking meters — While perhaps the scourge of the urban driver, parking meters ensure that spaces remain available throughout the day and generate revenue for cities and businesses. The New York City Comptroller (2019) reported that parking meters run by the city’s Department of Transportation generated over 200 million dollars in both 2017 and 2018. Furthermore, congestion was decreased by the presence of parking meters at their inception (Rich & Associates Marketing, 2021). Despite these benefits, most parking meters still have significant downsides.

Over recent years, parking meters have become a source of complaint for vehicle owners. Often, parking lots and street-side parking have one parking meter per floor or street, meaning the car owner must waste time to locate and receive their receipt (GMC, 2022). Often, these single-street, single-floor meters are unmaintained and dirty, not to mention that most parking meters only accept coins, which is inconvenient in the modern world. Additionally, the car owner is limited to preset amounts of time, meaning people must either overpay or risk being fined if they do not leave on time or are delayed (Eveleth, 2013). Even meters that accept card payments often have a mandatory minimum charge or sometimes glitch and stop working. For example, an incident occurred in 2020 where various parking meters stopped accepting cards (GMC, 2022). Despite these glitches, vehicle owners are still expected to pay by any means. 

As a more concrete example, a specific candidate for improvement is the parking meter system of the New Rochelle parking infrastructure. When a driver parks in one of the New Rochelle garages, they must make a note of their license plate and enter it into a parking meter. This type of meter may be present at one central location for smaller lots or with at least one on each floor of more oversized parking garages. However, through personal experience, the buttons on these machines are known to be consistently glitchy and insensitive to the user’s touch, making it difficult to use unless they already possess a third-party app that charges a premium for its use (City of New Rochelle, n.d.). If users lack mobile data connection or battery charge, they are forced to use the nonresponsive machines. Furthermore, the privatization of parking meters can lead to issues years later. The city of Chicago, Illinois, sold control of their parking meters to a private company in 2009. Parking rates quadrupled immediately, with fewer free parking days as well (Amariei, 2023). As the author notes, the company has already made an estimated 500 million dollars in profit, with 60 years left on their contract. The combination of inefficient privatization and faulty technological integration results in a parking system that is less effective and frustrating for users.

We propose to use a new system for cataloging and charging drivers for their parking space; we want to use the license recognition technology currently used for highway tolls to automate and smooth out the process of applying parking fees.

In the US today, automated license plate readers can be found on bridges, highways, street poles, lights., and even in police cars. The high-speed camera systems capture license plate numbers along with time, date, and location data and send them to a central server (Lynch et al., 2020). This data is then used to charge the toll to whoever that vehicle is registered under or even determine what vehicles were present at the scene of a crime (Lynch et al., 2020).

This system of license plate scanning and charging the registered owner is simple, automated, and ideally suited to our parking garage dilemma. Multiple cameras would be placed, one at each entrance and one at each exit, to scan cars as they pass through the range of vision. The video feed would then be processed to compare each vehicle’s time of entry to its exit time for the appropriate charge. With this system in place, scattered, independent, manual parking meter stations would be rendered obsolete in favor of one automatic, centralized system. We propose implementing this system into the New Rochelle parking lots, but others have already considered this method. Taiwan was the first to implement this in the city of Taipei. 

Our purpose behind this proposal is to find a way to make paid parking systems less stressful for everyday drivers just trying to get places, and we think our system will not only do that but also eliminate parking tickets, reduce errors, and make the entire process more efficient for everyone.

Objectives 

This project aims to assemble and implement a camera-server system using the cameras shown in Appendix B to automate parking fees for a public parking lot. The objectives of this project are therefore:

1.1 Design and assemble a prototype system that will:

-recognize and read all license plates upon entry and exit.

-catalog their information in a saved server.

-compare data and put out the relevant charge to the car’s owner.

-report errors and delete extraneous data.

1.2 Implement said prototype into a standard New Rochelle parking lot for several weeks to determine if this system can be more efficient and result in simultaneously more accurate charges to drivers with fewer errors than a standard parking meter system.  

Preliminary Literature Review 

There is much research surrounding this sort of system that already exists. Highway license recognition cameras are no longer new technology, and papers about the types of effective cameras, license recognition software and how it works, data storage, and server calculation can be found in abundance. To top things off, Chien et al. (2020) detail implementing a system similar to our proposition in Taipei, which has been very successful; this is good news for our proposal. 

Jan et al. (2023) performed research for an optimized smart parking meter system using a combination of camera and radar technology. The researchers noticed that smart parking meters that scan license plates are less accurate in foggy, rainy, or shadowy conditions than in clear conditions. This deficiency of the parking meter presents an obstacle to enforcing parking fares. Therefore, Jan et al. (2023) conceived and enacted tests for a new sensory radar that combines the technologies of RGB cameras and mmWave radars for more accurate scans. Their research concluded that this heterogeneous fusion of technology had an average accuracy of 99.33%. A huge step up from standard scanners. In our proposed New Rochelle parking garage parking system, accuracy, especially in conditions of low visibility, is critical to success. We plan to utilize the discovery made by Jan et al. (2023) to boost our license plate recognition system’s efficiency and functionality further.

In 2022, a paper by Al-Batat et al. laid out the design of an automated license plate recognition (ALPR) system. This system can process images to recognize vehicles, the license plate of that vehicle, and the text on the license plate. Initially, the algorithm processes the camera feed to determine whether or not a car is in the frame. If a vehicle is recognized, the algorithm records a patch of the image that includes only the relevant vehicles and can recognize multiple vehicles in one image. The vehicle patch is then processed to generate a patch with just the license plate. These patches take up less space than complete image files and can be processed separately without affecting the performance of the algorithm’s previous step. Finally, the algorithm parses the actual characters on the license plate. 

This algorithm can be implemented in our parking system. The authors designed a system capable of recognizing license plates on multi-lane highways, which would have significantly greater high-speed traffic than our proposed implementation. When processing an image with one vehicle, the authors recorded an average total processing time of .0549 seconds (Al-Batat et al., 2022). As most parking structures limit drivers’ speed, the entry and exit rate is far too low to overtake this time. With a powerful enough GPU, our computer system should be more than capable of processing the license plates of every vehicle that enters and exits a parking structure. The authors of this paper also included data that simulated uneven lighting and shadows when teaching their algorithm, and their methods were capable of processing the license plates regardless (Al-Batat et al., 2022). It is much easier to control the lighting for any covered parking structure than for a camera near a highway. As this system can handle edge cases with many variables, it can process images from an image source with fewer variables.

Other helpful information can be found in research done in China. Due to the rapid urbanization of China and the sudden increase in car owners across the country, there was a need to develop and integrate an intelligent transportation system (ITS) (Wang et al., 2023). Intelligent transportation systems aim to implement different forms of advanced science and technology to increase transportation efficiency (Wang et al., 2023). Data is collected and organized between vehicles, users, and the roads and then digested by artificial intelligence/machine learning algorithms to manage traffic efficiently to accomplish this goal. Features collected as data include license plate numbers, vehicle types, and speeds. Wang et al. (2023) aim to establish an algorithm to store and efficiently process all this information and components. Since license plate recognition is a crucial component of our established system, this paper provides meaningful recommendations for an algorithm that can be adapted to serve our purposes. 

The authors begin to outline the algorithm by establishing a traffic checkpoint where the vital information is recorded and stored (Wang et al., 2023). In our case, this checkpoint would be the entrance of the parking lot. The algorithm must store the vehicle’s properties as data in a time series (Wang et al., 2023). Essentially, the data is assigned a time stamp and organized by its license plate number. A jump hash consistency algorithm defines this first part of the algorithm (Wang et al., 2023). The information is then stored in an HBase, a columnar storage system used by large data platforms such as Google (Wang et al., 2023). In the second portion, a row-key generation algorithm must be used to assign a timestamp when the vehicle passes the checkpoint (Wang et al., 2023). This information is stored in regions and then divided into regional servers where the data can then be retrieved for use. Two algorithms are used for this process: a data distribution range generation algorithm and a multi-threaded sliding window for parallel request algorithm (Wang et al., 2023). Name Node servers and Data Node servers are the hardware needed to implement this algorithm. Name node servers will require more memory and less storage, while Data servers require more storage and less memory; 8G memory to 200 G storage for Name servers and 4G memory and 500G storage for data servers (Wang et al., 2023). This proposed algorithm is currently being applied in the “Tianjin Sino-Singapore Eco-city High-Performance Big Data Platform project” (Wang et al., 2023, p. 15). We plan to use a similar system in our proposal. However, because we require less data to be collected, we anticipate hardware specifications to be significantly reduced and the algorithms to be simplified. 

In Taiwan, a common practice used by all the on-street parking management authorities is that parking attendants must patrol the streets to issue parking bills. Parking attendants must also circle the roads every 30 minutes to charge them. Many city governments have initiated the testing of smart on-street parking systems since 2017. Taipei City has conducted trials using two types of sensors: wireless magnetometers and image recognition-based smart parking meters. When a vehicle occupies a particular parking space, and the wireless magnetometer detects the change in occupancy, the sensor transmits the event to the parking database through the NB-IoT interface (Chien et al., 2020). Subsequently, a parking attendant is notified and dispatched to the parking space to issue a parking bill. If a parking spot is equipped with smart parking meters, parking attendants are no longer required to issue bills because the license plate number is recognized and charged automatically. This means drivers can pay using smart cards at the meters or through the ParkingLotApp smartphone application (Chien et al., 2020). It is a convenient way to pay for parking and eliminates the need for physical bills and human interaction. Their goal is that if vehicles with Bluetooth beacon transmitters and on-street parking spaces have Bluetooth beacon receivers installed, the system can detect when a vehicle enters or leaves a parking space. The Bluetooth beacon technology is based on Bluetooth Low Energy (BLE) physical interfaces, meaning that the beacon transmitters can run on coin batteries and last several years with a long duty cycle (Chien et al., 2020). When a vehicle enters a parking space, the beacon receivers will detect its beacon packets containing a specific identifier. The beacon packets’ Received Signal Strength Indication (RSSI) will be processed using the Kalman Filter and then sent to the gateway (Chien et al., 2020). By comparing the measurements from the beacon receivers, the gateway can determine the occupancy of the parking space used by this vehicle and send the information to the parking management system to count its duration of stay (Chien et al., 2020). This is like our idea, but instead of using Bluetooth, we have cameras that would save the cars’ license plates, send them to the server, and then charge them for the corresponding time they have been in the parking by charging their cards directly. This can be applied to a wide range of vehicles without requiring specialized equipment in each car, unlike the need for Bluetooth beacons.

Technical Description of Process

Our setup for the camera apparatus is outwardly simple. Six CCTV cameras like the ones shown in Appendix B, each with a connected radar, will be attached to upside-down mounts, with each set of three being bolted to the upper lip of the garage’s entry and exit way. As shown in Figure 1, those cameras will be connected to the processing server, where their data will be fed to the recognition algorithm. The algorithm will then store and process the important data correctly. The specific methods will be detailed below. 

I. Server

All the information collected by our camera would first need to be stored in a data server. For this server, we anticipate needing 350 G of storage and 4 G of memory/RAM. For our name server, which will deal with our jump hash algorithm for scanning and recognizing license plates, recording charging rates, and billing them out, we will require more memory/RAM to operate at maximum efficiency. This name server will need only 150 G of storage but 6 G of memory/RAM. This server would also be responsible for automating the manual conversion between the color camera sensor and the IR sensor at night when less light is available.

II. General structure

Of course, the camera itself needs a mount and a place to be positioned very well. Most CCTV cameras come with their mounts, but it would make the camera too far away to see most license plates. We are proposing a mount like the wall mount bracket for dome cameras at the entrance. Of course, this will change depending on how and where the entrance to the parking garage is positioned, but this is the general idea. There will be two camera sets, one for each exit and entrance, which are often next to each other. We will also synchronize the time setting on them, allowing the cameras to record, determine the license plate number, and send the correct time and information without overloading a single camera. Because most CCTV cameras come with their own lighting and will each be equipped with a mmWave radar, we do not have to worry about illumination, and both cameras will be in the parking garage to avoid sun glare or random object interference. Cameras similar to those shown in Appendix B would be ideal for our implementation.

III. Algorithm

The automated license plate recognition (ALPR) system will primarily be adapted from Al-Batat et al. (2022). We will first develop software that processes the video captured by the cameras like the algorithm detailed in Al-Batat et al. (2022). As the article details, a patch from a frame of video can be captured when the algorithm recognizes a car. The entrance and exit lane cameras will each have a server to capture this initial patch along with the time of entry and exit, and a third server will parse the plate number and compare the times for entry and exit. This code will then have a controlled testing phase with simulated weather and lighting conditions. Once this testing is complete, the camera and server hardware can be installed in an actual parking garage. Once installed, a public testing phase will begin. As Figure 1 shows, if no number can be recognized, the computer will record that an error has occurred, and the image can be sent for manual processing by a city employee. The rate of errors will be necessary for analyzing how well this initial implementation performs in a parking garage. We plan to test this system for one month, with the entire project’s timeline shown in Appendix A. The billing can use existing systems, with the parking ticket system being an ideal system to expand on. 

Budget

The budget for this project is split into two sections, as shown in Table 1. First, we have personnel costs. A software development firm will need to be contracted to create the software that processes the images, as well as test and maintain the software over the course of the trial period. As Appendix A shows, they will be contracted for an estimated 286 hours, with the total cost of that time shown in Table 1. In order to ensure the smooth operation of the project, a contract attorney will also be hired for a flat fee. Next are the hardware costs. The cost of the three servers may increase if the proposed model is unable to run this software effectively. However, our estimation indicates that it will be sufficient. The mmWave radars and Cameras each have a flat cost, and there is an estimated rate for the installation of the cameras.

Table 1

Project Budget Breakdown

Budget Item	Cost Calculation	Item Cost
Personnel
Software Development firm	$82-$137/hour with est. 286 hours of work	$23,452-$39,182
Contract Attorney	$608 standard flat fee for contract review	$608
Implementation
Servers	$799/server for three servers	$2397
mmWave Radars	$18.24/radar with 6 radars	$109
Cameras & Camera installation	$200/camera and $125-$450/installation with 6 cameras	$1950-$3900
Total Cost		$28,516-$46,196