Skip to content
A Keras implementation of dynamic programming for shortest path tracking
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.gitignore
Dynamic Programming - Weld Path.ipynb
LICENSE
LaserLineTrack.cs
MultiTrack.cs
README.md
laserline.png
weld1.png
weld2.png
weld2proc.png
weldProcPy.png
weldTemp.png

README.md

LaserLineTracking

A Keras implementation of dynamic programming for shortest cost line path tracking.

It is possible to implement dynamic programming in Keras layers to take advantage of GPU acceleration for laser line tracking.

Laser tracking is important in a wide variety of applications: seam tracking (for automated welding robots), self driving vehicles (to determine distances and road shape), 3D scanning and others.

Below is a sample of the results (see .ipynb notebook):

Original image

Processed image

C# and OpenCL

If you are interested in a complete version of this code using C# and OpenCL feel free to browse through the .cs classes.

To do:

(I have this code implemented in C/C# and OpenCL for a different application. I will port the code to Python if and when the need arises or if someone requests the features).

[ ] Embed backtracking in Keras

[ ] Carry information from more than 1 pixel above/below (influences the max slope of the tracked line - for 1 pixel the max slope is 45 degrees. For 2 pixels it increases to 63 degrees)

[ ] Compute the best N lines from the backtracked points

[ ] Track two or more lines simultaneously.

You can’t perform that action at this time.