This project involves the work I did for [removed name for confidentiality, calling it ORG from now]. The end product was a robust MATLAB-based tool to automatically generate UAV trajectories. The areas that my project touched: operations research, algorithms, image processing, interfacing. The tool is now being used by scientists at [ORG] for less computationally expensive simulations. Software used: MATLAB and STK by AGI.
The problem:
- Currently, in most places including [ORG], UAV trajectories are NOT mapped automatically before hand and multiple simulations are performed to find a path that avoids enemy radars (modelled as Domes). Thus, there is a need to find a more accurate and computationally inexpensive solution for the same.
- The second issue is that there exists no functional tool for linking MATLAB/other programming suites and the simulations software - STK. There is a plugin offered by AGI, which doesn't seem to work in our case.
My approach (now a working solution):
- Use the geo-map from the simulations software STK, and set up the scenario with starting point, ending point, and dome-shaped radars.
- Import the image into MATLAB and use image pre-processing algorithms so that the A* algorithm for cost-optimization can be applied.
- Use color codes for costs and color thresholding algorithms to differentiate between regions that a UAV can and cannot go through.
- Find the required lat-long coordinates in MATLAB
- IMPORTANT STK has a provision to take HTML commands for initializing path points for a UAV. We make use of this exploit and generate HTML commands as printed text, from the MATLAB application.
- These commands are then fed to STK directly (simply copy-pasting) and we have thus solved the problem.
Please feel free to write to me at sammit.bitspilani@gmail.com for any queries regarding this. Some pointers for using this repository:
(@) Currently working approach:
- Use STK to generate snap of the scenario to be evaluated. Use .tif for better results
- Remember to fill the area targets with one colour and remove everything else from the scenario.
- Run preprocess2.m in MATLAB and choose required colours. (Google impixel)
- Run ip_approach1.m and feed it the path of the processed image.
(@) You'll get the plot in MATLAB, use photoshop to overlay it onto original image.
(@) Currently working on conversion of waypoints according to scale and feeding them to STK
EDIT:
(@) Conversion algorithm added. This is essentially how it works: top_left = [42.64595 36.08668]; %Got this from STK cursor bottom_right = [19.18550 110.76360]; %Got this from STK cursor
converted_points = [bottom_right(1)+(size(rad_map2,1)-OptimalPath(:,1)).*(top_left(1)-bottom_right(1))/size(rad_map2,1) top_left(2)+(bottom_right(2)-top_left(2)).*OptimalPath(:,2)/size(rad_map2,2)];
Simple explanation: Use the top-left and bottom right lat-long coordinates to suitably scale and translate the coordinates. NOTE: Images have their origin on the top left and coordinates traditionally start from bottom left (hence latitude needed to be subtracted to invert it, in a sense)
(@) Algorithm added to generate STK connect commands Generating a text file called STK_instr.txt with looped waypoint generation commands Result from this text file is to be fed to the HTML API of STK