Principal Investigator

Alex Hauptmann

Status

Completed

Start Date

Jan. 1, 2015

End Date

Nov. 30, 2017

Project Type

Research Advanced

Grant Program

Private Funding

Grant Cycle

2016 Traffic21

Visibility

Public

Abstract

The project will develop a pilot application of smart parking in the long-term parking lot. The research project is designed to develop methods for optimization and ease of passenger convenience home-to parking-to building. This involves using camera observation to identify open and available parking spaces, keeping track of the spaces in a database in terms of which ones are as available, reserved or taken. With this information The project will develop an app for smart phones which can be activated when entering the lot, the car is then tracked by cameras upon entry, and combined with GPS the app guides the driver to the most optimal open spot. The research project goal is ensuring success regarding data, time, and value, eliminating stress and promoting convenience.    

Description

The project has the following research goals in order to create a mechanism that allows users/drivers to reach their parking spots easily. 1. Use cameras installed in the airport parking lot to navigate drivers to specified parking spot. 2.  Manage the parking spot data in a convenient way. The airport smart parking study will collect video from up to 8 cameras installed on 4 lamp post (2 cameras per post) in a section of the long term parking lot at Pittsburgh International Airport. The airport already has signs posted indicating the video surveillance is in effect. Data collection is expected on not always consecutive days, where days with different weather patterns are of prime research interest (cloudy, sunny, rainy, foggy, snowy, windy, stormy). Videos will be collected on a computer located in the parking office at the airport, which will also be the location of the database that tracks spaces and monitors cars. The video data will be analyzed to develop computer algorithms for automated analysis of empty parking spaces and tracking of vehicles across a network of cameras under different environmental conditions. This project seeks to improve the customer parking experience and utilization of available parking spaces. Our core vision for improved parking assistance is to monitor vehicles coming in and out to be able to direct drivers to available parking spots using large public displays or a smart phone app. We will develop methods for monitoring available spaces in the airport parking lot. Once the system knows where a parking space is available, it can communicate this information to central display screens at the parking lot entrance or to apps on individual smart phones which can guide drivers directly to the nearest available space. The project initially will assess accuracy of different parking space monitoring approaches under different conditions (visibility, temperatures and precipitation), as well as establishing a life cycle costs, including estimates of maintenance expenses and durability of the sensors. To minimize requirements for installation of new wires, each sensor will wirelessly communicate with a base station to update the parking space occupancy information. Concurrently we will build a system for displaying available parking spaces on a map and a navigational guidance app for smart-phones. The objectives are as follows 1. An area within the long-term parking lot at Pittsburgh International Airport will be instrumented for monitoring. 2. Eight Camera’s will be mounted on 4 of the light towers (2 cameras per tower) to visually track vehicles as they enter the lot. 3. Video technology tracks the car as it parks in the stall. A server in the parking lot office will securely store and manage all the data. 4. An app will be created to test guiding the driver to the closest parking spot. The app can get activated by the driver before the car enters the airport parking lot. Camera tracking and GPS will then navigate the driver to the selected parking spot. Preliminary feasibility studies have been successfully conducted using data from other video, since administrative issues prevented us from getting video data directly from the airport.
? Work conducted in the first phase of this project included successful test of recording from IP camera networks that use power-over-ethernet (POE) instead of requiring their own power sources. We also attached this to the type of optical network currently installed at the Pittsburgh airport.
? We have developed a camera calibration technique to rectify images of wide angle cameras in order to allow a direct mapping to 3-D reconstruction of the visible scene.
? We have created preliminary car tracking approaches and empty parking spot detectors.
? We collected basic geography information about airport parking lot including latitude and longitude of important points at key locations on map. The proposed new research phase of this proposal has the following goals:

? We will incorporate an algorithm into our system to calculate the shortest route between two geo points on the map. General road data will be provided by Google Maps, and a newly developed filter algorithm will be used toselect road points in a database and identify interest points along the route. All selected points can be linked and shown as the the route on the screen.
? We will be building a first version of a system to: (1) Build a connection between the GPS sensor and the android app. (2) Estimation and filter for GPS locations. (3) Devise a simulation algorithm for the motion of user vehicles. (4) Provide step by step navigation instructions to the users through the app. (5) Provide directions. The basic function of the navigation instructions include: (1) telling users how many feet are left to drive in their current direction. (2) telling users the exact number of feet before they have to make a right/left turn and (3) warning users when they have to change direction.
? We will create an online data base which contains all parking spot information at the airport. The database will store information about each parking spot such as the section, the row number, column number within the row, latitude and longitude of the parking spot and the availability of the spot (taken, available or reserved for a car being guided there at the present).
? We will creat an interactive interface for users so that they can choose the most suitable parking spot for them and reserve it. Also, this interface can be used by airport parking management to help in managing and improving parking utilization. Video  monitoring research. The advantage of video monitoring lies in the fact that it can be installed or maintained without interfering with the day-today operations of the facility. Since cameras will be deployed on poles, there is a significantly reduced risk of tampering or incidental damage. A computer vision based system can also easily scale to parking areas of arbitrary sizes and can be easily re-calibrated if the layout of the parking spot layout changes for any reason. Computer vision based systems can offer more than just occupancy related information. By using advanced classification techniques we would be able to classify vehicles into types and identify vehicle color. An image recording of every vehicle in every spot together with the occupancy period can also be maintained.   We will place cameras at each light pole in the parking lot, which will provide a complete view of all the parking spaces between light poles. The camera will connect to a computer system that detects and tracks cars and identifies the filled and empty spots based on prior calibration. Since light poles only have power during nighttime hours, we will co-locate a battery with the camera to provide separate power for daylight operations. Using computer vision approaches based on interest points based methods and image subtraction, we will be able to detect differences in parking space occupation. On initial install, the system will need configuration and calibration for modeling the parking area per camera view. Detecting differences in the occupancy can be done by simply subtracting one image from the other. We utilize the constraint that the parking lot is planar, which allows us to use planar homographies to perform a warping of the different raised and angled camera views to the actual map of parking spaces on the ground. This allows a unique label to be assigned to every parking spot. If cameras overlap, we use the computed homography to warp the view of a parking space from one camera image to identify it as identical to the parking space in the warped image viewed from the other camera position. Drawbacks: It is not clear to us if adverse weather conditions significantly will significantly affect the performance of the video monitoring approach. Based on prior research, our solution will be to model the background as a slow, time-varying image sequence, which allows the system to adapt to changes in lighting and weather conditions. Continuous vehicle tracking entry to exit.  In addition to the monitoring of the empty and occupied spaces, the cameras will be tracking each car as they enter the lot at the barrier. In conjunction with license plate recognition (which is commercial technology not part of this proposal), this system would identify each car in each space by its license plate. We have previously developed a system called the Marauder’s Map that performs robust person localization and tracking in real world surveillance scenarios, especially in complex environments with occlusions, long view angles and relatively sparse surveillance camera coverage. We will adapt this method to a tracking-by-detection approach for vehicles. Given vehicle detections from the cameras viewing the entrance gates, our framework can take advantage of all important cues such as vehicle detection information (as distinguished from persons, bags, strollers, etc.), vehicle color, vehicle type and non-background information to perform tracking. In the case of partial or full occlusions of vehicles, learning approaches are used to uncover the manifold structure in the appearance space with spatio-temporal constraints. Our approach also enforces a mutual exclusion constraint, which guarantees each vehicle detection will be assigned to exactly one individual vehicle track.

Timeline

We expect the research phase of this project to last two years. Month 1 -6: Complete instrumentation of one section of the long-term airport parking lot (8 cameras) Month 7 -12: Analyze and evaluate collected data, adjusting instrumentation as appropriate Month 13-18: Develop database to track large scale space monitoring and app for drivers Month 19 – 24: Improve accuracy of algorithms and develop real-time monitoring and tracking systems.

Strategic Description / RD&T

Deployment Plan

If successful, the project will or spin-off a company to apply this approach to the whole airport parking lot. The technology will then be suitable for licensing to any parking management organization with large outdoor parking areas. There is a potentially huge market around the world at airports, arenas,
stadiums or concert venues.

Expected Outcomes/Impacts

• Data for tracking cars through the parking lot will be collected and potentially used for general multi-camera traffic research. We expect to collect over 6 months of daily parking lot observation data, • System success will be measured in terms of accuracy (missed detections, false alarms) of identifying empty parking spaces. • We will also measure the accuracy of tracking cars across the whole parking lot in terms of correct tracking and missed tracks. • A database and visualization interface will be developed to manage the availability of spots and assign spots to new entries. • An app will be developed to help drivers find the best available parking spot.

Expected Outputs

TRID

Individuals Involved

Email	Name	Affiliation	Role	Position
sudevbohra@cmu.edu	Bohra, Sudev	CSD	Other	Student - Undergrad
alex@cs.cmu.edu	Hauptmann, Alex	CSD/LTI	PI	Faculty - Research/Systems
hlu2@andrew.cmu.edu	Lu, Han	LTI	Other	Student - Masters

Budget

Amount of UTC Funds Awarded

$75000.00

Total Project Budget (from all funding sources)

$75000.00

Documents

Type	Name	Uploaded
Publication	1811.11969.pdf	Feb. 19, 2019, 4:42 a.m.
Publication	CADP_IEEE_Camera_Ready_Final.pdf	Feb. 19, 2019, 4:43 a.m.
Final Report	35_-_airport.parking-final-report.pdf	March 18, 2019, 5:30 a.m.
Presentation	35_-_UTC_Powerpoint_Airport.pptx	March 19, 2019, 5:29 a.m.
Poster	35_-_poster.ppt	March 19, 2019, 5:30 a.m.
Publication	Automatic vacant parking places management system using multicamera vehicle detection	Feb. 28, 2021, 5:55 a.m.

Match Sources

No match sources!

Partners

No partners!