Group 10 Pull Request

[ccardo]

Key Additions

• Autopilot Management: Added a new executable script ap_restart.sh designed to quickly restart the autopilot system. This script is found inside the root folder. It replaces several commands that our group had to run in the terminal, due to a failure in the RTP viewer initialization during Flight UDP Paparazzi runs. The issues were always resolved by this quick AP restart, but the root cause for these issues was not addressed. This script represents a quick solution to the problem. This script should not be used in-flight. Further instructions can be found inside the .sh file.

• New Features & Configuration: Setting Sliders: Integrated UI sliders into orange_avoider.c to allow for real-time parameter tuning:

  1. Speed Multiplier: Configurable range within [0.1; 2],
  2. Heading Increment Setting: Configurable range within [1°; 45°].
  

• Data Labeling & Validation Suite: Mask Painter GUI (mask_painter.py): An AI-assisted graphical tool that simplifies the manual mask creation process.

  1. Function: Allows users to "paint" binary masks directly over images.
  2. Controls: Supports Left Mouse Button (LMB) for painting, Right Mouse Button (RMB) for erasing, and mouse wheel/Middle Mouse Button (MMB) for zoom and pan.
  3. Validation Scripts: Provides multi-panel visual feedback comparing original images, raw classifier outputs, and filtered results (including edge density and bounding box clustering).
  

This last feature is part of a larger addition by our team. Inside sw/tools/dataset_mask_generator/, there are Python scripts that can automate the generation of a large binary mask dataset, to identify the ground, poles and plants. The folder also includes some sample masks, hand-painted by the user, which are used to train a ML model which then can generate similar masks for an arbitrary number of drone images. The full pipeline is reported inside README files in the folder itself.

Documentation & Code Clarity

1. Improved Commenting: Added comprehensive comments to orange_avoider.c/.h and cv_detect_color_object.c/.h.

2. XML Integration: Added specific pointers within header files to the relevant settings locations in the project's XML configuration.

3. README Updates: Included detailed documentation for the masks_data folder and a project-wide README explaining the labeling pipeline and computational requirements.

Notes

1. Training Challenges: masks have been kept for unsuccessful training attempts, for plant vases (due to small pixel size) and gates (failed improvement attempt). These are preserved for archival and future research.

2. Deployment: Note that while the labeling pipeline is highly effective for data preparation, it currently has high computational requirements that prevent direct real-time use on the drone hardware. This way, it can only be used for its intended purpose (i.e. creating large mask datasets) and cannot be exploited for actual on-board computer vision.