How to spot empty parking spaces?

Question

I want to do a final year B.Sc project on parking space detection. Can anybody give me some link related to it?

• Any text-book, tutorial or anything?
• What would be prerequisite for this project?
• What type of skills(programming/math) are needed?
• What are the initial steps to do?
• What type of readings(algorithms of image processing) are needed?

Detail added in comments: i'm going to use camera, not infrared. i would like to use still images, or one camera which captures images from a parking lot. i think reak-time processing will be tough, at this moment i just need to start the project. so still images will work fine. but later it may b a real-time project

Answer 1

The Computer Vision Homepage: Established at Carnegie Mellon University in 1994 to provide a central location for World Wide Web links relating to computer vision research.

Answer 2

Here is a few links on image processing that I could find that could be relevant to your situation.

• http://www.efg2.com/Lab/Library/ImageProcessing/Algorithms.htm
• http://scialert.net/fulltext/?doi=itj.2009.114.127
• http://www.rob.cs.tu-bs.de/en/research/projects/parkspot

If you want, you can redo my search by clicking here.

EDIT: Now that I look at it, you might also be considering magnetic car detection in your parking place. If the budget is not a problem, I'm sure there is already some commercialized system that would be able to be installed to detect car in specific spots. Probably more expensive than cameras however.

Magnetic car sensor may vary from 1$ to 6$ each depending on what you need. So if you have a 200 places parking lot... I would see a magnetic detector (like they have at red lights) as a very feasible and cheap solution.

Answer 3

Steps for spotting empty parking spaces from a video:

• Rent a sweat-shop in China
• Hire 100 Chinese people, equip each with a work station.
• For every 5 minute video segment, select a Chinese person in random and send him the video
• Make some fancy GUI for marking empty parking spots.
• Process the the output from the GUI and present it to the user.
• Profit.

Skills needed: Ability to run a sweat-shop.
No reading required.

Also notice this solution is very scalable.

Answer 4

Assuming that you are looking for parking meter spaces...

I would attempt to get the public data, if available, for the geo-locations of parking meters for a city/town. I would then have a GUI to refine and verify the locations on satellite map (google maps perhaps). You would also need access to live satelite images. Work on detection of background w/o car versus with a car in the spot. Hand test your algorithm until it is perfect.....get gray hairs and pull them out. :)

Answer 5

Are you placing a camera over the spaces or like a security camera which can scan larger areas?

Language: You could write this in C++ or C#. Up to you.

Multiple Cameras Over Large Scan Area Method

Here is a research paper using large scan areas. I am not confident that this method will work well in real-time based on occlusions and difficulties with correlating multiple cameras over a parking lot. If the parking lot is small, this is easy, but it gets more difficult with having to merge multiple areas to get an overall count.

Camera over each lane method:

My recommendation is to use a simple D-Link Internet Camera (DCS-3220 or DCS-3220G) with Motion Detection. They are not expensive. You will have to write a server component that will be able to receive the images. The D-Link Internet Cameras have the capability to FTP the images to a pre-defined directory. If using multiple cameras, have them all point to the same FTP and just a different directory such as C:\Garage\Space1, C:\Garage\Space2, etc. When motion is detected in the image, you can have it FTP the image to your server. Since you did not say you are using loop detection or something else, you'll have to use special algorithms to determine if a car, motorcycle, or truck is actually in the space.

** With traffic loop method ** : If you ARE using loop detection, then this makes it much easier because then you can determine if a big hunk of metal is in the "lane" (i.e space). The D-Link Internet Cameras have Digital I/O so you can 'trigger' the camera to take the picture and send it to your server. I have done this in the past for many applications and it works great. You will need to either have a separate controller to handle the loop detection or iterate over all of your 'space' directories to see if an image is in there. The latter method is slower because you could have a garage with over a 1,000 spaces and the hard drive would be cranking!

** No traffic loop method: **

Next, your server component will have to determine if a car is actually in the space. This is the hard part! You will need to go to the Computer Vision Homepage and find sample projects and research papers which have already covered this in detail. This is the best I can give you.

Having said all that, cameras over the spaces will not be cheap!

Answer 6

Oddly enough, my first work mentor was given a very similar problem when he started out. He was asked to do a cost/benefit analysis of a system to scan the parking lot for cars with their lights on, read the license plate and send email to the owner to let them know they need to turn the lights off.

The result of his analysis (at the time) was that it would be cheaper to hire someone to walk the parking lot with a hammer and break all the headlights that were on and then buy replacements for the employees.

You could use a similar system.

Answer 7

Check out AForge.NET, it's an OpenSource C#.NET project with lots of code demonstrating many of the things you're looking for.