User Tools
Differences
This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision | |||
|
cs:fusion:start [2019/04/16 21:52] Andrew Rink |
cs:fusion:start [2019/04/16 21:54] (current) Andrew Rink [How it Works] |
||
|---|---|---|---|
| Line 20: | Line 20: | ||
| ===== How it Works ===== | ===== How it Works ===== | ||
| - | The basic concept of the node is simple: take the bounding box points provided by the SLAM system and attach the labels from the object detection system so we get an identified object in our map. Both SLAM and object detection treat the cameras as 2D images, so by calculating the closest x,y coordinate pairs within a single image we can reasonably determine which labels belong to which bounding boxes. Since SLAM provides both a 3D vector to the bounding box and the x,y coordinate within the image, the end result is a 3D vector to a labeled object in the map. The final step of the fusion node is to transform the object from its current frame to the "world" frame and publish that to /tf. | + | The basic concept of the node is simple: take the bounding box points provided by the SLAM system and attach the labels from the object detection system so we get an identified object in our map. Both SLAM and object detection treat the cameras as 2D images when finding objects, so by calculating the closest x,y coordinate pairs within a single image we can reasonably determine which labels belong to which bounding boxes. Since SLAM provides both a 3D vector to the bounding box and the x,y coordinate within the image, the end result is a 3D vector to a labeled object in the map. The final step of the fusion node is to transform the object from its current frame to the "world" frame and publish that to /tf. |
