5.2.1 Background

A lot of cities worldwide are working to become smart. One of the most popular applications in smart cities is the smart parking system, as they help people to save time, reduce fuel consumption, and decrease carbon dioxide emissions. Smart parking systems have a defined architecture with specified components. Although there are many kinds of research done, still a small amount of work has been done for the concept of outdoor smart parking systems in Sri Lanka.

Furthermore, most of these parking systems do not cover the following areas: real time availability of slots in different weather conditions, navigating user to the available slot by showing the position with great accuracy and the facility of the fully automated parking payment system by detecting vehicle number plate in any condition and by providing payment patterns. Most importantly a lot of systems do not have a unique market value because most parking systems only provide the parking facility and do not focus on providing some benefits for the users who come to the parking area, like digital targeted advertising.

Many researchers have successfully implemented many algorithms to detect objects. But a small amount of work has been done to identify vehicle occupancy detection in the outdoor parking area under various conditions such as flexibility under different lighting varieties and meteorological conditions which will provide efficiency and high accuracy for real-time delivery.

Also, there are many algorithms and methods to correct GPS errors because when more precise positions require, errors in GPS data need to be corrected. Thus, positioning accuracy is a challenging issue for location-based applications, but a small amount of work has been done in using map-matching algorithms to correct these GPS errors that provide more accurate results.

Furthermore, this smart parking system is to develop User Targeting Techniques that is the process of predicting the relevance of advertisements. Many industries are sending fascinating models to make target crowd bunches for different commercials. The work in this area includes Analysis using Machine Learning and it is capable of identifying and displaying the relevant content of a particular object.

Image processing is used to recognize vehicles by identifying license plates. It is basically used for parking access control and security purposes. First, the vehicle comes near the entrance gate and a signal is sent to the camera. Then the camera is activated and takes a snapshot.

The system applies certain algorithms to analyze the vehicle image. Next, the image is enhanced, the vehicle plate is placed, and the character is extracted from the vehicle plate. The system attempts to coordinate the perceived vehicle plate number to the vehicle license plate database. If “Allow”, the gate will open, then vehicles will be allowed through.

Recognition of vehicle license plates also has the advantage of storing images of vehicles that can be used to fight crime. Vehicle speed, weather conditions, vehicle type, distance between vehicle license plate and camera, plate type (rectangle, bent type), orientation of vehicle license plate, vehicle font character type, efficiency, and accuracy are variables that affect the recognition system [19].

License plates usually have a rectangular shape with a known aspect ratio, so they can be extracted by finding all possible rectangles in the image [18]. To find these rectangles, an edge detection method is commonly used.

Hence, a literature review is carried out based on every sub topic.

5.2.2 Review of Literature

“Smart parking reservation system using short message services (SMS)” by Hanif, Badiozaman and Daud [5] have presented a smart parking system which reduces the time wastage during finding a parking spot in commercial car park areas [5]. This parking reservation framework is inherent such a way that clients can book their parking spots via a SMS. The system has developed as a fully automated system by using a Peripheral Interface Controller (PIC). This microcontroller is able to store data about available parking spaces, include passwords and permit or deny access to the parking area. A prototype of a car park system was designed to show the potential of the proposed work. The demonstration has shown the system’s ability to reserve the car park, gain entry to the parking area and thus remove the hassle of looking for empty parking lots. There are many advantages in this system. Such as; enhance security with password requirements and with the ease of use, the system can be used and applied anywhere. And there are some disadvantages either, like; High installation cost, GSM function creates a bottleneck and the microcontroller needs to put a lot of lads that can disturb the system.

Yang, Portilla, and Riesgo et al. stated that “Smart Parking Service based on Wireless Sensor Networks” [6] and, the authors have presented the design and implementation of a Smart Parking Services prototype system based on Wireless Sensor Networks (WSNs) that allows drivers to effectively locate the available parking spaces [6]. This system consists of wireless sensor networks, embedded web server, central web server, and mobile phone applications. In the system, low-cost wireless sensor network modules are installed in each parking slot equipped with one sensor node within the device. The parking space status is detected by the sensor node and is periodically reported to the embedded web server via the wireless sensor networks that are deployed. Using Wi-Fi networks in real time, this information is sent to the central web server and the vehicle driver can locate empty parking lots using regular mobile devices too. The android app is easy to use and with a great interface, GPS helps cover the maximum area available and it displays various options for parking can be listed as advantages of this project, while the user cannot use the reservation function, unavailability of multi-layer parking in infrastructure can be listed as disadvantages.

Shih and Tsai et al. stated that “A Convenient Vision Based System for Automatic Detection of Parking Spaces in Indoor Parking Lots Using Wide-Angle Cameras” [7]. They have proposed an indoor vision-based parking system. The wide-angle fisheye-lens or cat dioptric cameras which are easy to set up by a user with no technical background were proposed in this research. Easiness of the framework arrangement mostly originates from the utilization of another camera model that can be aligned utilizing just one space line without knowing its position and heading. Also the camera allows convenient changes in detected parking space boundaries. A genuine vision-based parking area has been built up and important examinations directed. The accuracy, practicality, and strength of the proposed techniques were showed by experimental results. Although it has disadvantages like; No mean is provided to address weather conditions that may affect visibility and Reservations. General users can easily set up the system without any technical background and wide angle camera is used to cover the entire territory of the parking area can be categorized under advantages [7].

“Automated Parking System with Bluetooth access” by Singh, Anand, and Sharma [8] have developed a fully automated parking system for both two wheelers and cars. This described system extemporizes upon the current parking framework by improving its security includes and mechanizing the parking process accordingly wiping out the requirement for manual mediation. The system has an inbuilt Bluetooth reader to authenticate and identify the owner. This kills the utilization of tokens or paper bills. Space the executives and mechanization are performed with the assistance of an ARM microcontroller which controls the mechanical engines to leave the vehicle at a fitting parking area [8].

“Car Park Management with Networked Wireless Sensors and Active RFID” by El Mouatezbillah, Djenouri, Boulkaboul and Bagula [9] have proposed an automatic car park management. Integration of networked sensor/actuator and radio frequency identification (RFID) technologies was used and the emerging internet of things (IoT) context was used as service provider. In view of this combination, they have proposed a scalable and low-cost car parking framework (CPF). A fundamental model usage and experimentation of certain modules of the proposed CPF has been performed. Trial results exhibit extensive decrease in cost and energy consumption [9].

“Automatic Parking Management System and Parking Fee Collection Based on Number Plate Recognition” by Rashid and Musa [10] have talked about an automatic parking system and electronic parking fee collection method based on vehicle number plate recognition. The objective of this system was to create and execute an automatic parking system that would increase the effficiency and security of the public parking lot as well as collecting parking fees without using a magnetic card. This system is developed with less human interaction. Hence it was a big help to develop the iParking system which is a fully automated smart parking system. iParking system contains fully automated parking guidance system as in this research. In this research they have used image processing in parking operation and payment procedure, and in iParking system machine learning will be used to minimize the errors of image processing [10].

“New Smart Parking System Based on Resource Allocation and Reservations” by Geng and Cassandras [11] have suggested a new method of “smart parking” for a city environment. The system allots and saves an ideal parking space dependent on the cost capacity of the driver joining the vicinity to the objective and the expense of parking. The system approach solves the problem of mixed-integer linear programming (MILP) at every point of the decision specified in a time-driven sequence. Based on simulation results, compared with uncontrolled parking processes, this system reduces the time wastage on finding a parking space, whereas the overall parking capacity is more efficiently deployed. They have also described full execution in a carport to test this system, where another light system conspires is proposed to ensure client reservations [11].

“An Accurate GPS-IMU/DR Data Fusion Method for Driverless Car Based on a Set of Predictive Models and Grid Constraints” by Wang, Deng and Yin [12] have proposed an accurate GPS–inertial measurement unit, data fusion method in view of a lot of prescient models and inhabitance framework requirements.

In this paper, the system proposes a novel data fusion method for precise localization problems of driverless cars using a set of ARMA predictive models and occupancy grid constraints. It is only based on on-board GPS– IMU and DR navigation data [12].

Study on “Differential GPS Positioning Methods” by Shao and Sui [21] have presented about the three typical methods for differential GPS according to the different type of information sent by the base station, they are differential position; differential pseudo-range; differential carrier phase [21].

From this research the main idea about GPS positioning was gained [21].

“Automatic parking space detection system” by Bibi and Nadeem [3] have presented a vision based smart parking framework to assist the drivers to find an appropriate parking space and book it efficiently. In here they have used image processing and though this research image processing procedure can be understood. When the car enters or exits the parking lot, its presence and details like size of the vehicle take into an account and those information will be used in further process [3].

“Parking spaces modeling for inter spaces occlusion handling” by Masmoudi and Wali [2] suggested an idea to provide smart parking performance and to solve the problem which caused due to the lack of space in parking areas and also between two vehicles. They used an event-based detection approach to detect objects and tracked them using a Kalman filter. The goal of this paper is to introduce a new approach to provide parking vacancies with reliable results and to address the problem of interplace occlusion [2]. This research paper is proposing a new approach for the parking architecture extraction in view of the on-road surfaces, as it is the less impeded model.

Then arrangement of the cars in the parking area, mentioned parking spaces on street surfaces shown in Figure 5.2 with red area, will reflect the parking model with the less occlusion possible and provide good performance for the detection of parking events [2].

Fabian et al. stated that “An algorithm for parking lot occupation detection” [4], and he has proposed an unsupervised vision based system for parking space detection. But they have used image processing to work under many different conditions. This system claims that the major problem in images detection is the occlusions and shadows using recognition algorithms [4].

“Parking lots space detection” by Wu and Zhang [13] have proposed a parking detection system which has the following steps:

1. pre-processing
2. extraction of model features
3. multi-class SVM (Support Vector Machines) recognition
4. MRF (Markov random field)-based verification [13].

Initially, the input video outlines are turned to uniform axes and portioned into little fixes containing 3 parking spaces such as Figure 5.3. This paper presented a novel method via using only a few frames which are captured by a single camera for unsupervised parking lot space detection [13].

“Parking space classification using convolutional neural networks” by Cazamias and Marek [18] have described a parking occupancy classifier. In this research they have used convolutional neural network (CNN) as a classifier. The workhorse of this algorithm is a Convolutional Neural Network (CNN), a deep learning data structure that works especially well for computer vision tasks [18].

“Networked Wireless Sensors, Active RFID, and Handheld Devices for Modern Car Park Management” by Djenouri, Karbab, Boulkaboul and Bagula [19] have implemented a Car Parking Framework (CPF) which can manage a smart parking system that combines sensors (detection of parking spaces), RFID tags, and readers [19]. This system uses serial cable communication.

Sharma, Kulkarni, Kalbande, Dholay et al. stated that “Cost Optimized Hybrid System in Digital Advertising using Machine Learning” [20], but they have suggested an Audience Targeting Strategies method of predicting the importance of ads to individuals. Machine learning algorithms analyze the personal data of individuals that they share with companies on a voluntary basis. The results are promising for targeting the audience [20, 21]. The authors have suggested many deep neural network techniques.

Audience Expansion: Audience Targeting Techniques is the method of assessing the importance of ads to individuals. Several industry giants, such as Facebook and LinkedIn, are introducing innovative models to build target audience audiences for different advertising. Research in this area includes Advertisement Analysis using Machine Learning, Audience Expansion for Online Social Network Advertising [20].

“AI-Based Targeted Advertising System” by Yu, Lee, Lim, and Razak [20] have proposed an intelligent targeted advertising system with following functionalities,

1. Gender and age identification using facial recognition
2. Vehicle recognition
3. Multiple object detection.

Facial recognition is applied to recognize gender and age on the basis of facial characteristics. Various item acknowledgment innovation is utilized for vehicle types and for various classes of article acknowledgment dependent on their novel attributes [20].

In this research machine learning techniques have used same as iParking to improve the effectiveness. Machine learning algorithms are implemented for higher acknowledgment exactness and subsequently accomplished better focused on publicizing impacts [20].

The models used in the recognition framework are pre-prepared with the idea of machine learning for exceptionally precise outcomes and improved execution. With the system’s ability to display targeted advertising content, advertisers benefit as they could significantly reduce their promotional costs due to the effectiveness of targeted advertising. As for the viewer, they will be introduced to material that is more important to them and will be given the products and services that they may need [20].

5.6.1 Lot Availability and Occupancy Detection

Firstly, the video image of the parking area was acquired by a fixed camera, and then the pre-processing of each frame image was carried out, including image color conversion, image de-noising, and image morphological transformation [4]. The processing of the image was divided into system initialization, image acquisition, image segmentation, image enhancement, and image detection. The thresholding method was used to get a free image of darkness and the Canny Edge detection algorithm was suitable for edge detection to separate the shadow and the object. There were some identified filters to solve these various detection issues under certain weather conditions and lightning varieties, such as HOG (Histogram of Oriented Gradients), SIFT (Scale-invariant feature transform) and color Spaces (YUV, HSV, and YCrCb). After the image processing stage, the occupancy detection classifier status was given using CNNs (Convolutional Neural Networks) and a binary SVM (Support Vector Machine).

5.6.2 Error Analysis for GPS (Global Positioning System)

The accurate path to the available slot was calculated by using the Global Positioning System. Thus, this system used a method to enhance the position accuracy of a GPS receiver which has used map data rather than any additional sensors. A Map-Matching (MM) algorithm was used to position data from a navigation system with spatial road network data. The points extracted from the roads prior to operating the algorithm were utilized as map data. A MM algorithm uses a range of positioning and navigation data, including position, distance, speed, and road network topology, to identify the correct road segment on which the vehicle is traveling and the location of the vehicle on that road segment. The key task of the MM algorithm was to identify the correct road segment from a pool of candidate road segments. The method matched the disparity between the trajectory of the GPS and the digital map information.

5.6.3 Vehicle License Plate Detection System

The first step in the license plate recognition system (LPR) was to find the position of the plate from the vehicle image. The system also recognizes the vehicle license plate at the entrance gate and can take action to enter or leave the vehicle. A typical PC-based system with a PIC microcontroller and a video camera catches and processes video frames containing a visible car license plate.

An automatic license plate recognition technology consists of three basic steps. They are recognition of the position of the plate attached to the vehicle, text image segmentation and convert character images into beautiful identifiable images. The plating area usually consists of a white background and black text. Therefore, the transition between black and white is very intensive in this area. Finding the area that contains most of the transition points is enough to find the plate area. For this purpose, this system applied a canny edge detection operator to the vehicle image to obtain transition points.

In this process, the blob (Binary Large Object) coloring algorithm has a powerful architecture for determining closed non-contact areas in binary images. In this task, the System applied a 4-way blob coloring algorithm to a binary-coded license plate image to retrieve characters. After implementation, the segmented characters were obtained from an image of the license plate area. This system also suggested ANN (artificial neural network) for OCR (Optical Character Recognition). ANN is a statistical model of a real-world system. Simplify the code and improve the quality of the image recognition process.

The system uses ultrasonic sensors to confirm the parking or leaving of the vehicle. Then the payment will be automatically calculated according to the vehicle type and the time period it stayed in the park and parking fees will also be applied according to usage at peak times.

5.6.4 Analyze Differential Parking Behaviors and Pricing

Discovered the pattern of behavior and its predictability of parking lot occupancy/reserved and parking time period. K-means cluster algorithm was used to partition the vehicles in a specific time, after this system calculates the features for each vehicle to the parking lot. According to the other periodicity, identify what are the peak hours and then we identify the main frequencies and used these main frequencies to recreate the time domain sequence and forecast the future value. According to the above predictions rating was changing with the time periods. If it was peak hour, parking rate was getting high. If it was normal hours, parking rate was not changing and if many parking slots were free, the parking rate was getting low. Likewise, parking rates will be changed.

5.6.5 Targeted Digital Advertising

Machine learning algorithms can be utilized from numerous points of view to accomplish amazing customized experience. The three most famous ways are prescient investigation, cost advancing and crowd focusing on. Advanced Advertising has caught wide consideration from the market. It is an integral asset to contact right individuals at the right time. Additionally, it lessens the expense of broadcasting ads as the ad is shown uniquely to individuals who may be keen on the substance. The interest prediction for audience targeting provides accuracy using the Naive Bayes classifier. The process started with data collection. After a period of time, the system collected data for each advertisement. Audience categorization was then carried out to classify users in categories then customize the ad for that class is shown to specific users. This was filtered by users Gender, Age, Job, etc.

Using a Bayesian algorithm, it then compared the information of users with that of the advertising database, figured out the most suitable ads and compared the results, and showed the ad to the user.

5.6.6 Used Technologies

The following set of technology domains were used in this module for implementation of the system.

• Computer Vision

• • Vacant parking lots detection

• Machine Learning

• • To train a dataset of parking slots in order to recognize them and for classification of parking availability and occupancy based on visual features extracted through Computer Vision.
  • Position accuracy improvement of GPS.
  • Improve Audience categorization for digital targeted advertising.

• Deep Learning

• • A branch of Deep Learning, which is Convolutional Neural Networks (CNN), will be used to preprocess images and achieve performance.
  • Artificial Neural Networking (ANN), will be used to character segmentation of vehicle license plate.

5.6.7 Specific Tools and Libraries

The following list provides the specific tools and libraries that were employed in the system module.

• Android Studio
  Frontend web framework in designing the UI for the module
• OpenCV
  Programming function Library which is mainly used in real-time computer vision.
• MATLAB
  Use to process images one generally writes function files, or script files to perform the operations

Use for GPS position correction

• Java—Create Admin console
• React—Development of mobile application
• Pdfkit
  Python library for generating PDF documents
• Tensorflow
  TensorFlow is a free and open-source programming library for dataflow and differentiable programming over a scope of assignments. It is a representative numerical library and is utilized for AI applications, for example neural systems.

Table 5.1 shows the tools that used in the research according to the fields and the particular areas.