Perception branch - #16
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 7b492998c9
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| detection_msg.results.append(hypothesis) | ||
|
|
||
| # 보조 정보: echo로 확인하기 쉽게 JSON 저장 | ||
| detection_msg.id = json.dumps({ |
There was a problem hiding this comment.
Detection2D.id stable or empty
When these raw YOLO detections are fed into cctv_to_map, det.id is interpreted as a tracking ID: non-empty, non-numeric IDs are hashed, while empty IDs fall back to the detection index and skip velocity estimation. This JSON includes bbox coordinates and confidence, so it changes from frame to frame for the same object; downstream objects therefore get unstable IDs and churn velocity state instead of degrading as untracked detections.
Useful? React with 👍 / 👎.
| self.image_sub = self.create_subscription( | ||
| Image, | ||
| self.image_topic, | ||
| '/camera2/image_raw', | ||
| self.image_callback, | ||
| 10, | ||
| 10 | ||
| ) |
There was a problem hiding this comment.
The launch files still pass config/yolo_detector.yaml, and the docs describe overriding image_topic for additional cameras, but this node no longer declares or reads those parameters. In any launch or CLI run that sets image_topic, detections_topic, or annotated_topic for another camera, the node still subscribes to /camera2/image_raw and publishes on the fixed /yolo/* topics, which breaks the documented multi-camera path.
Useful? React with 👍 / 👎.
| hypothesis.pose.pose.position.x = float(anchor_u) | ||
| hypothesis.pose.pose.position.y = float(anchor_v) |
There was a problem hiding this comment.
Writing the corrected anchor only into hypothesis.pose.pose.position does not affect the current CCTV-to-map pipeline: repo-wide search shows cam_to_map_node and cctv_to_map_node still project det.bbox.center + size_y / 2 and do not read this pose field. As a result, the new person/box anchor logic is ignored by the homography path, so boxes still map from the bbox bottom rather than the intended ground-center point.
Useful? React with 👍 / 👎.
2 participants
No description provided.