11/06/2024 Wednesday Meeting Notes

[himanshunaidu]

The following notes are the progress and next steps for each sub-task.

1. Computer Vision Model Update

Assigned to Himanshu.
Will be picking up the frame control in the coming weekend. I anticipate taking care of a few other issues as noted here.
#21
#80

2. Segmentation Mask Post-Processing

Assigned to Ihor and Dmytro.

a. Object Tracking
Ihor is looking through the OASIS code for object tracking. The OASIS code utilized a version of ORB tracking. Implementing this on Swift will take time.
While homography transformation can be done, a brute force approach with smoothing could work.
Having a region boundary and the centroid for an object, across frames can help with an initial version of object tracking.

IMU sensors can be utilized. Thanks to the presence of IMU sensors, we do not have to estimate change, a task which ends up involving rigid affine transformation. With IMU sensors, we have r-vectors as well as angular movement.
IMU sensors in iPhones can also give us confidence intervals (need to check which ones do).

IMU drift. While frame-to-frame, this will not be an issue, we are currently capturing frames through manual user interaction. This will add significant error.

Other keywords: Kalman filter, SLAM

Conclusion: Try a brute force approach first, and then identify if we need fancier methods.
The following are the pending tasks:

• Utilize a brute force approach that involves averaging, smoothing and low-pass filtering
• Research on the data given by iPhone IMU sensors
• Check if the IMU sensors give us confidence intervals

b. Union of Masks
Dmytro is still working on this.

3. API Transmission

GoInfoGame currently uses the Overpass request manager, and OSM-based web auth.
We import data from OSM as well as TDEI. But we focus on sidewalks, and we want to update in bulk, so we just push to TDEI.
Another reason for using our own workspace is the issue with ID updates. Whenever someone updates an object in OSM, the ID of that object changes, which is not favorable. TDEI avoids that.

Currently, the GoInfoGame developers have updated documentation, so we will be going through the updated documentation, and come back with any possible questions we come up with.

We do not have access to upload databases to TDEI workspaces, so we are supposed to get access from Professor.

Mariana does not have access to some of the documents, so we will have to work on getting her access to the relevant documents listed in the following issue:
#61 (comment)

4. Annotation View Feature Improvement

This work is currently contingent upon the work of Segmentation Mask Post-Processing and API Transmission.
It hasn't been assigned yet, but Himanshu or Mariana will most likely be dealing with it.

[himanshunaidu]

Next Immediate Steps:

1. Get access from Professor to be able to Upload Databases to TDEI workspaces
2. Carry on with the pending tasks discussed with the Professor on object tracking