Fix SRT parsing and GPS altitude handling

[kemaluzel]

Summary

Fixes issues in SRT parsing that caused incorrect GPS altitude handling.

Context

This PR is based on the discussion and findings here:
https://community.opendronemap.org/t/several-issues-found-in-srt-parsing-and-gps-altitude-handling/25790

Changes

• Improved SRT timestamp parsing
• Corrected GPS altitude extraction and handling
• Added safer checks for missing or malformed SRT values

[smathermather]

Thanks for this @kemaluzel, and thanks for the forum post. This looks to make a big difference for high precision workflows with video. Did you use LLMs / code agents to write this?

[kemaluzel]

Thanks! I used chatgpt as an ai assistant for refactoring and sanity-checking parts of the code. The issue analysis, debugging process, dataset testing, and the RTK/SRT logic corrections are based on my own workflow and field-generated data. Since I generate my own SRT files from RTK GNSS receivers, the fixes are grounded in real high-precision datasets.

[smathermather]

Thanks for the additional info. For clarity, your ChatGPT prompt was what you posted on the forum and then you wrote the pull request based on that exchange?

[kemaluzel]

For some time, I had suspected there was an issue in the video + SRT processing workflow. Over the past year, I tested multiple datasets and repeatedly tried to correctly match the SRT data with the video, but the processed outputs were consistently deformed in some way.

To better understand the issue, I decided to review the related code. Since I am not very familiar with Python, I analyzed how the parsers handle the SRT structure based on the SRT files I generate. With the help of ChatGPT, I identified potential problems in the altitude parsing logic and some regex definitions.

I shared my findings with the community as soon as possible. To prepare the pull request efficiently, I did use ChatGPT to help draft and structure the patch. However, the issue identification and data analysis were based on my own real-world datasets and testing.

I have not yet had the chance to fully test the corrected code. If this contribution helps improve ODM even a little, I would be very happy.