Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 11

PARKING SPOT

UTILIZATION
DETECTION
By: Stephan Huber
Why?
Save time
Less
frustration
Happier
people
Image Process Flow
Cascade Classifier

Uses pretrained model to locate


object in image
Requires 2000+ negatives and
1000+ positives
Processed16 stages
3+ hours processing time
Returns bounding box points
Which Bounding Box?

Immediately remove any too far (above horizon)


Sort them by highest y-value
Closest should be best match
Who Should Connect?

Mask area to the right (right side)


Mask area to the left (left side)
Also constrain the distance 20px>Box>85px
Is It Empty?

Split ellipse in half


Calculate histogram on each side
Each only has 3 bins and 3 values
Compare histograms
Image sections with similar
intensities will match
Results!
Mediocre results
Found 80% of line endings
Missed 6 of 40
Large number of false positives
Cause many incorrect pairing
Processes around 45 fps
Single core 2.5GHz
Bounding boxes change 45 fps

Professor
Chuah
Going Forward
Discussion Improve image detection
Hold bounding boxes
Improve histogram comparison
Dynamic ellipse width
Pros
Fast enough for real time
Cons
No counting
Missed 6 of 40
Large number of false positives
Boxes move too fast
Processes around 45 fps
Single core 2.5GHz
Bounding boxes change 45 fps
More Results!
Early Methods
Canny Edges
Hough Lines

You might also like