340 likes | 723 Views
SDP ’07 Team Frasier. Automated Parking Lot Attendant. Tom Cleary Matt Regan. Bill Ryan Adam Bailin. Current System. Disorderly Confusing Antiquated. Large Parking Lots. The larger the parking lot, the more difficult it is to find a parking space. Choosing a Lot. Many obstacles
E N D
SDP ’07 Team Frasier Automated Parking Lot Attendant Tom Cleary Matt Regan Bill Ryan Adam Bailin
Current System • Disorderly • Confusing • Antiquated
Large Parking Lots • The larger the parking lot, the more difficult it is to find a parking space
Choosing a Lot • Many obstacles • Parking spaces are obscured • Hard to map This is a bad lot
The lot we chose • Fewer obstacles • Parking spaces easily identifiable • Easier to map This is a good lot
However • Still things we need to worry about
The Camera Axis 210 Network Camera Set up on 2nd Floor KEB
System Overview Project all about image processing Two main parts: control and processing Need a central way to control system Basic steps of control system 1) Take picture 2) Send to Matlab 3) Receive from Matlab 4) Display to user
Controlling the System Block diagram for control system Generate readable result Initialize system Wait for timer to expire Query camera, grab snapshot Display to user Save snapshot locally with unique filename Send image data to Matlab for image processing Matlab returns processing results
Take picture every 3 seconds using Timer Run m-file from Matlab Wait for Matlab to return results Arrange results in human-readable form Create image – layout of parking lot with indication as to which spots are taken Controlling the System
Software Using Microsoft’s .NET framework Classes WebRequest() - request web resource (image.jpg) HttpWebResponse() – returns jpg data stream FileStream() – saves stream locally Timer() - take pictures at interval
User Interface • Will present user with computer generated map of parking lot
Problems and Solutions Learning curve for Visual Studio and MATLAB Network congestion (wireless vs. wired) .jpg image size (640x480) Delays to/from Matlab
Must read picture into Matlab “imread(‘c:\snapshot.jpg’)” Image is 3-dimensional(red, green, blue) Snapshot 480x640x3 uint8 We have our picture on file, now what?
Our image Processing Basic idea: Image Differencing! Is the new snapshot different from the base snapshot? If so, something must have changed Cut large snapshot into smaller pieces Each small piece is of one parking spot Pixels are manually mapped to each spot All processing done on small pictures individually
Scaling Example • This is one example of pixel mapping • Most processing will be done on these small pictures
How Different? No two pictures are alike Glare, shadows, random ambiences. How different are two pictures? Correlation coefficient! Variable which represents how different or alike two pictures are Between -1 and 1, 1 being two identical pictures A correlation coefficient below the threshold causes concern! State of parking spot is changed New snapshot becomes the base
A visual.. t0…… t5….. t10… • A visual of how the program will run
Differencing Issues Ambience's blocking camera position What if a truck blocks the view? Solution! Timing buffer The base picture is only changed if the new picture is different for a time Something that is blocking the camera will likely move away
More Issues… Cars aren’t the only thing that can cause a change Daylight gradually changes the new snapshot from the base Solution! Use full snapshot A subtraction will show where the most change took place
Determine Ambient Conditions Look at area of just pavement If average of pixels is similar, spot is probably empty
MDR Specifications Mount camera in good location overlooking a lot near Knowles Engineering Building and connect to network Able to import an image into an image processing program Able to manipulate an image using basic image processing techniques
Live view of camera http://abyss.ecs.umass.edu:8080
Images RGB Grayscale Edge Detection
Images Picture 1 Picture 2 (Picture 1) – (Picture 2)
Looking ahead… Need to explore the effects of weather conditions such as rain and snow May need to consider alternate image processing solutions due to the following observations: Pixel subtraction may not be accurate based on time of day Obstructions (groups of people, cars driving through parking lot) Glare on window directly in front of camera – solved with box Have many ways of determining spots – can average them, have threshold for ‘spot taken’ event We’re over the learning curve Our demo