ScaleDigitizer

The goal of this project is to extract digits from the seven digit display of a bathroom scale to get the displayed weight. Having multiple videos available, one can average the weight and save this to a database to track relative weight changes from week to week.

Requirements

For running the project as is, following packages are needed:

Python 3
Numpy
openCV
skimage
imutils

What does it do?

We need videos of the bathroom scale, such as the following example:

We then process each image individually. First off, the region of interest(ROI) of the scale will be calculated by applying a blur, edge detection and hough line transformation. The result is a cropped image which then is taken under a 4-point transformation and a threshold, resulting in following output:

Next on, morphological dilation is applied to the threshold image to thicken the digits and seperate them via openCV contour methods. The result can be seen here:

By dividing the segments into sub-segments for a seven segment display of each digit, one can simply count the number of pixel in relation the the total count of pixels of a sub-segment to find out whether the seven-segment-part is activated or not. This way, a recognition of digits is possible. Recognized digits can be displayed:

Finally, a filtering of the recognized values is applied: the most occurrences of recognized digits are taken as the recognized value for every video. These votings are converted to a single floating point number and then all recognized numbers are averaged over the number of available videos inside the designated folder.

Every average is saved to a database and compared to the previous entry to calculate a relative difference in weight.

Using the script

python analyizeScale -video_path=PATH_TO_VIDEOS -database_name=NAME_OF_DATABASE -visualize=BOOL

Arguments are optionals, since we will look for videos under ./videos/ and the database will be stored under ./database/data by default. Results will not be visualized be default.

References

https://www.pyimagesearch.com/2017/02/13/recognizing-digits-with-opencv-and-python/

https://github.com/suyashkumar/seven-segment-ocr

https://www.kurokesu.com/main/2017/02/20/dumb-thermometer-gets-digital-output/

https://github.com/kazmiekr/GasPumpOCR.git

(Medium: https://hackernoon.com/building-a-gas-pump-scanner-with-opencv-python-ios-116fe6c9ae8b)

https://github.com/bikz05/digit-recognition

https://github.com/g-sari/pyautodigits

https://towardsdatascience.com/scanned-digits-recognition-using-k-nearest-neighbor-k-nn-d1a1528f0dea

Special thanks