![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3747125%2F926edd59b0d9a7d92f8ee5a459cb4e05%2Fxy_plot_ep_translations.png?generation=1572756254621734&alt=media\"){kind=spacer}
Tried plotting X,Y values from ego pose translations. Plot seems to indicate camera front left , front and front right are all in exactly the same center line tracing the direction of the car. I was expecting left and right to be offset from the front.
\n\n
Notebook showing the plot. Is there a gap in understanding? Do the translations also have to be rotated to get the real world X,Y coordinates of each sensor.
", "messages": [ { "id": "664038", "postDate": "11/03/2019 04:45:41", "content": "Tried plotting X,Y values from ego pose translations. Plot seems to indicate camera front left , front and front right are all in exactly the same center line tracing the direction of the car. I was expecting left and right to be offset from the front.
\n\n
Notebook showing the plot. Is there a gap in understanding? Do the translations also have to be rotated to get the real world X,Y coordinates of each sensor.
", "rawMarkdown": "Tried plotting X,Y values from ego pose translations. Plot seems to indicate camera front left , front and front right are all in exactly the same center line tracing the direction of the car. I was expecting left and right to be offset from the front.\n\n\n\n[Notebook](https://www.kaggle.com/mpdroid/question-on-ego-pose-translations) showing the plot. Is there a gap in understanding? Do the translations also have to be rotated to get the real world X,Y coordinates of each sensor.", "votes": null }, { "id": "664151", "postDate": "11/03/2019 08:41:08", "content": "You gotta rotate them after translation.
", "rawMarkdown": "You gotta rotate them after translation.", "votes": null }, { "id": "664310", "postDate": "11/03/2019 13:31:42", "content": "ok figured out the correct transforms I think. Is this correct?\n\nsensor_coords_in_world = np.add(\n ego_pose_translation,
\n Quaternion(ego_pose_rotation)\n .rotate(calibrated_sensor_translation) )\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3747125%2Fa54f85e6a42d3d16f68da63fd495879c%2Fxy_plot_sensors.png?generation=1572805718402098&alt=media\")
I got same result but sensor_vector for lidar front right and left need be with sign minus. Just need time for experiments to figure out why.
\n\n![\"\"](\"https://www.kaggleusercontent.com/kf/22945358/eyJhbGciOiJkaXIiLCJlbmMiOiJBMTI4Q0JDLUhTMjU2In0..xL-mNlUkxv2YH5fi1ZLo3g.GglRN8SV0u0Ext6RWFvDVrnIRY1eLirU_HErqeQxAZ0IFBrI01oJff0CAYxob8Z7__S1-AyOgJDB62FY6LGep6LZk1ANrHXI9LQ02bAfsb8DE-dk1Fiwqc5TO_0BE4HHCb54dlc4vJ_Y0JlBIVQFipc6d0LDs_-5MMwKnLGYC51Gh_2fvJgULiUPGajL9nCh.f-WGOBuGDUmM01DJpA4Vsw/__results___files/__results___14_0.png\")
sample_cam_token = my_sample['data']['CAM_FRONT']
\n\ncam = lyft_dataset.get('sample_data', sample_cam_token)
\n\ncs_record = lyft_dataset.get('calibrated_sensor', cam['calibrated_sensor_token'])
\n\nsensor_vector = np.dot(Quaternion(poserecord[\"rotation\"]).rotation_matrix, cs_record[\"translation\"])
\n\nposerecord = lyft_dataset.get(\"ego_pose\", cam[\"ego_pose_token\"])
\n\nx.append(poserecord[\"translation\"][0] + sensor_vector[0])\n y.append(poserecord[\"translation\"][1] + sensor_vector[1])
", "rawMarkdown": "I got same result but sensor_vector for lidar front right and left need be with sign minus. Just need time for experiments to figure out why. \n\n\n\n sample_cam_token = my_sample['data']['CAM_FRONT']\n \n cam = lyft_dataset.get('sample_data', sample_cam_token)\n \n cs_record = lyft_dataset.get('calibrated_sensor', cam['calibrated_sensor_token'])\n\n sensor_vector = np.dot(Quaternion(poserecord[\"rotation\"]).rotation_matrix, cs_record[\"translation\"])\n \n poserecord = lyft_dataset.get(\"ego_pose\", cam[\"ego_pose_token\"])\n \n x.append(poserecord[\"translation\"][0] + sensor_vector[0])\n y.append(poserecord[\"translation\"][1] + sensor_vector[1])", "votes": null } ], "comments": [ { "id": 664151, "author_name": "rishabhiitbhu", "author_url": "", "post_date": "11/03/2019 08:41:08", "content": "You gotta rotate them after translation.
", "votes": null, "replies": [] }, { "id": 664310, "author_name": "mpdroid", "author_url": "", "post_date": "11/03/2019 13:31:42", "content": "ok figured out the correct transforms I think. Is this correct?\n\nsensor_coords_in_world = np.add(\n ego_pose_translation,
\n Quaternion(ego_pose_rotation)\n .rotate(calibrated_sensor_translation) )\n
I got same result but sensor_vector for lidar front right and left need be with sign minus. Just need time for experiments to figure out why.
\n\n
sample_cam_token = my_sample['data']['CAM_FRONT']
\n\ncam = lyft_dataset.get('sample_data', sample_cam_token)
\n\ncs_record = lyft_dataset.get('calibrated_sensor', cam['calibrated_sensor_token'])
\n\nsensor_vector = np.dot(Quaternion(poserecord[\"rotation\"]).rotation_matrix, cs_record[\"translation\"])
\n\nposerecord = lyft_dataset.get(\"ego_pose\", cam[\"ego_pose_token\"])
\n\nx.append(poserecord[\"translation\"][0] + sensor_vector[0])\n y.append(poserecord[\"translation\"][1] + sensor_vector[1])
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "664038": "Tried plotting X,Y values from ego pose translations. Plot seems to indicate camera front left , front and front right are all in exactly the same center line tracing the direction of the car. I was expecting left and right to be offset from the front.\n\n\n\n[Notebook](https://www.kaggle.com/mpdroid/question-on-ego-pose-translations) showing the plot. Is there a gap in understanding? Do the translations also have to be rotated to get the real world X,Y coordinates of each sensor.", "664151": "You gotta rotate them after translation.", "664310": "ok figured out the correct transforms I think. Is this correct?\n````\nsensor_coords_in_world = np.add(\n ego_pose_translation, \n Quaternion(ego_pose_rotation)\n .rotate(calibrated_sensor_translation) )\n````\n\n", "664505": "I got same result but sensor_vector for lidar front right and left need be with sign minus. Just need time for experiments to figure out why. \n\n\n\n sample_cam_token = my_sample['data']['CAM_FRONT']\n \n cam = lyft_dataset.get('sample_data', sample_cam_token)\n \n cs_record = lyft_dataset.get('calibrated_sensor', cam['calibrated_sensor_token'])\n\n sensor_vector = np.dot(Quaternion(poserecord[\"rotation\"]).rotation_matrix, cs_record[\"translation\"])\n \n poserecord = lyft_dataset.get(\"ego_pose\", cam[\"ego_pose_token\"])\n \n x.append(poserecord[\"translation\"][0] + sensor_vector[0])\n y.append(poserecord[\"translation\"][1] + sensor_vector[1])" }, "source": "meta" }