-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge pull request #409 from una-auxme/374-research-why-does-the-car-…
…hit-obstacles 374 Research: Why does the car hit obstacles
- Loading branch information
Showing
1 changed file
with
56 additions
and
0 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,56 @@ | ||
# Research: Why does the car hit obstacles | ||
|
||
## Summary of current obstacle detection in perception | ||
|
||
- [Object detection](#Object-detection) | ||
- [Distance calculation](#Distance-calculation) | ||
- [Publishing of Outputs](#Publishing-of-Outputs) | ||
|
||
For more details, please refer to the current documentation. | ||
|
||
## Object detection | ||
|
||
Active sensors: | ||
|
||
- Center camera | ||
- LIDAR | ||
|
||
Inactive sensors: | ||
|
||
- Cameras: right, left, back | ||
- Radar | ||
|
||
## Distance calculation | ||
|
||
- The vision_node receives depth-images from the lidar_distance_node for the specified camera angle | ||
|
||
- According to the distance-to-objects documentation, the LIDAR sensor's 3D values are projected onto a 2D image that matches the exact dimensions of the camera image. | ||
Numpy formulas are used to reconstruct the depth image, with distances in meters represented as pixel values (grayscale image). | ||
If no distance is found in the depth image, “infinity” is returned for this bounding box. | ||
|
||
- The LIDAR sensor may flicker; a higher spin rate of the LIDAR provides fully reconstructed depth images but with lower resolution for the LIDAR points | ||
|
||
## Publishing of Outputs | ||
|
||
- Class_index | ||
- Min_X | ||
- Min_Abs_y | ||
|
||
## Dependencies of detected and hit obstacles | ||
|
||
- Are all obstacles detected? No | ||
- Does successful object detection depend on: | ||
- Speed of the car? | ||
- Size of obstacles? | ||
- Type of obstacle? | ||
- Dynamic/static obstacle? | ||
|
||
-> So far, no dependencies have been found | ||
|
||
## Important aspects found during research | ||
|
||
- Sometimes objects are not detected (sometimes e.g.: cyclists, construction site signs, open car doors) | ||
- Most objects that are hit are detected | ||
- Sometimes distance calculation is incorrect (distance is set to infinity just before the collision) | ||
- Distance in x is alway > ~0.5 (is that a problem?), sometimes distance to a collided car is greater than the distance to a parked car on the side of the road | ||
- Significant issues in dead-end situations: -> 1.2 seconds of reversing -> the car often immediately collides with the obstacle again |