search

Greenstand / Greenstand-Overview

Tree Tracking Fighting Poverty and Climate Change - This repository contains Contributing, Project Overview, Roadmap, etc
https://www.greenstand.org
GNU Affero General Public License v3.0
45 stars 11 forks source link

Mobile optimization challenge - optimizing convergence time for accuracy across devices #125

Closed Davidezrajay closed 11 months ago

Davidezrajay commented 2 years ago

PROBLEM/OPPORTUNITY

The time it takes for the Mobile app to capture an image needs reducing without reducing quality of the GPS-based location data coming from the app.

We need you - a data analyst - to study and solve the problem...

OUTCOME:

  1. Data-driven decision on convergence time-out strategy
  2. Find ways to speed up a positive convergence on certain phones.
  3. Faster app on the ground and more impact.
  4. Suggestions for more detailed track info
  5. Open an ADR discussing strategies and consequences for adjusting the time-out on GPS convergence.

BACKGROUND:

The Android App is being used to track millions of trees.

Planters and organizations use the app to take repetitive photos of the trees as they grow.

TIME TO TRACK

The largest barrier to using the system is the TIME TO TRACK per tree (planters are tracking several hundred trees a day, for organizations with millions of trees, the seconds and costs add up quickly.

The fastest the app has ever been at a pre-tree level is ~3 seconds and current times are ~20 seconds.

The app can be tested by taking as many photos as possible in a short amount of time, for example, 7:20 seconds / 50 photos.

GPS LOCATION ACCURACY

The primary increase in TIME TO TRACK has come from major advances in location accuracy. The 3-second app has been notorious for sending useless location data making it nearly impossible to match tree captures over time.

The latest version of the app is slow, however, it is producing high-quality data. It relies on GPS convergence to pinpoint its location before allowing the user to make a capture of the tree. Relaxing GPS accuracy requirements are directly in conflict with data needs for capture matching.

RESEARCH STATMENTS:

We have data to study to ascertain: (1) Are we spending too much time converging on good devices (we are already converged, but still waiting) (2) Are we wasting time trying to converge on devices that just don't converge (wait for the timeout)

SUGGESTED STARTING POINT:

Find device info for all the tracking sessions. Find data sets related to the problem. Look at data sets, tracking time varies from 18 to 7 seconds between photos on different devices

Davidezrajay commented 11 months ago

Replaced by use of track file