Open himanshunaidu opened 4 days ago
Special Considerations:
Dmytro Suggestion: While the user is mapping a sidewalk, we take user location into account. We use the user location and consider only those image segments for which the user location changed. We ignore those segments for which the user location changed, assuming that these segments were just captured on the side while mapping sidewalks (like segment 1 in the image)
Assuming that the user is going to capture a footway (e.g. sidewalk) in a straight line, create an algorithm that utilizes the segmentation history to map out the footway.
Note 1: As the simplest approach possible, we may end up using only the center point for each segment to create center lines between two consecutive segments each. Suggested by @NaturalStupidlty But in this approach, how will we capture width information? Would we use a running mean? Discussion on this is welcome.
2. It is possible that the user does not map out a footway in a continuous straight line. They could capture some other image in another location in between the sidewalk mapping.
Note 2: Dmytro's suggestion may be used for this special consideration. This would however, lead to another edge case pointed out in the next Note.
3. Are we going to consider the following use case: While the user is mapping out a footway, in between, they end up capturing an image that contains another footway of the same class
Note 3: Down the line, this special consideration will especially become a problem with this approach.
We can use an approach with clusters of points to find the footways of the same type during different capturing scenarios while distinguishing different sidewalks. Every time when we consider a point in a footway we may include it in the footway or form another one based on its coordinates. If the point is far enough (depending on how far the user could move and the segmentation frame rate) we can just form another potential footway (and when or if it will have enough points to be considered a real footway, not a segmentation mistake we can send it to the database).
Collecting Sidewalk Network Data at Scale for Accessible Pedestrian Travel: https://dl.acm.org/doi/abs/10.1145/3441852.3476560
OASIS: Automated Assessment of Urban Pedestrian Paths at Scale: https://arxiv.org/pdf/2303.02287
Notes: https://www.notion.so/OASIS-Paper-analysis-00414a258c0547d8a1f1e47567866e71
Our app needs to enable users to map out footways such as sidewalks. Currently, we intend to be able to create center lines for each footway segment that is identified by the Computer Vision model. Based on the segment and its center line, we would also want to send data on the sidewalk width.