Skip to content

Feet Plus Shoe size Measurement using OpenCV without Deep Learning

License

Notifications You must be signed in to change notification settings

wildoctopus/FeetAndShoeMeasurement

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FeetAndShoeMeasurement - Feet and Shoe size measurements without Deep Learning

Problem : Determine the Feet and corresponding Shoe size of a person without Deep Learning, given a image clicked by the person itself.

How to Evaluate?

  • Clone the repo
  • Install required packages using requirements.txt
  • keep the image inside "data" folder. A sample image has already been kept in data folder.
  • run main.py

Working Aproach

  • convert raw image to HSV format
  • Remove noise using Gaussian Blur
  • Run k-means clustering on preprocessed image for color based segmentation
  • Detect the edges in clustered image.
  • Find contours in Edge Detection output
  • Generate the bounding Box to get the height/width of Paper and Feet

Libraries used

  • OpenCV
  • Sklearn

Assumptions

  • Printer Paper is used as a reference (Height/Width is known and White background will help in Preprocessing)
  • Foot should be in center, touching one edge of paper.
  • Floor color should be different than white.
  • Image should be clicked form top angle.
  • Paper should be completely visible in the clicked image.

Sample outputs

Shoe Size charts for Mens and Womens

Designed Solution gives feet size in CM's, that can be compared with below chart for Corresponding Shoe Size

Limitations

  • If floor color is white, then it will difficult to segment the paper.
  • Feet should not go out of the paper. Currently A4 size printer paper has ben chosen as reference. To fit all sizes, A1 paper can be taken as reference.
  • If there is no closed curv after edge detection, its difficult to find bounding box. Only closed curv is considered as a contour. And bounding box is generated on top of that. As it can be seen in the below image, bounding box is not generated on feet but only on the paper.

Other Ideas

  • Deep learning models (Trained on feet and shoe Dataset) can be used to segment the paper and feet with more accuracy. One can use "Part Grouping Network"

References