The GPS field presents a lot of domain knowledge that can appear daunting at first. If you're new to the field I would recommend looking at the problem in these tiers, in order of complexity and/or how much domain knowledge is required:
\n*derived.csv files. This list shouldn't be considered definitive, and we'd love to hear other ideas about how the machine learning community can get started most efficiently!
", "messages": [ { "id": "1304570", "postDate": "05/12/2021 17:47:27", "content": "The GPS field presents a lot of domain knowledge that can appear daunting at first. If you're new to the field I would recommend looking at the problem in these tiers, in order of complexity and/or how much domain knowledge is required:
\n*derived.csv files. This list shouldn't be considered definitive, and we'd love to hear other ideas about how the machine learning community can get started most efficiently!
", "rawMarkdown": "The GPS field presents a lot of domain knowledge that can appear daunting at first. If you're new to the field I would recommend looking at the problem in these tiers, in order of complexity and/or how much domain knowledge is required:\n\n1. Smoothing out the baseline estimates. This requires essentially no domain knowledge.\n2. Integrating readings from other phone instruments, like the accelerometer.\n3. Satellite triangulation using the `*derived.csv` files. \n4. Building triangulations directly from the raw gnss logs. This may require more time to get working, but should also offer more opportunities for error correction.\n5. Incorporating external data for controls like satellite readings from base stations in the area.\n\nThis list shouldn't be considered definitive, and we'd love to hear other ideas about how the machine learning community can get started most efficiently!", "votes": null }, { "id": "1304577", "postDate": "05/12/2021 17:51:52", "content": "thanks for the list @sohier! I started from 1. by applying the Kalman Filter approach naively. One more thought before trying out more advanced ideas: if GPS is used while driving, aligning the signal to the network of roads and preventing from jumping from one street to another street in no time should also improve accuracy.
", "rawMarkdown": "thanks for the list @sohier! I started from 1. by applying the Kalman Filter approach naively. One more thought before trying out more advanced ideas: if GPS is used while driving, aligning the signal to the network of roads and preventing from jumping from one street to another street in no time should also improve accuracy.", "votes": null }, { "id": "1306938", "postDate": "05/14/2021 06:42:42", "content": "Thanks for sharing the list @sohier! Ensembling models can also help here is my feeling. Fingers crossed 🙏
", "rawMarkdown": "Thanks for sharing the list @sohier! Ensembling models can also help here is my feeling. Fingers crossed 🙏", "votes": null }, { "id": "1322314", "postDate": "05/25/2021 11:08:55", "content": "What is \"smoothing\" in this context?
\nIt's not \"Label smoothing\" right?
GPS from a phone can be quite \"jittery\" (moves around quite a bit), so some simple smoothing/averaging (especially if the runs have more than one phone that you can combine) can be a pretty good start at getting a reasonable looking path.
", "rawMarkdown": "GPS from a phone can be quite \"jittery\" (moves around quite a bit), so some simple smoothing/averaging (especially if the runs have more than one phone that you can combine) can be a pretty good start at getting a reasonable looking path.", "votes": null }, { "id": "1331162", "postDate": "06/01/2021 10:02:10", "content": "thank you. i think its diffcult for me……
", "rawMarkdown": "thank you. i think its diffcult for me......", "votes": null }, { "id": "1333817", "postDate": "06/03/2021 04:43:11", "content": "Has anyone tried trilateration? I would like to know the results It produces.
\nI've tried it on my own, but I'm getting nan as resultant latitude and longitude after trilateration.
\nI've tried incorporating code from:
\nhttps://gis.stackexchange.com/questions/66/trilateration-using-3-latitude-longitude-points-and-3-distances/415#415
Can you share something for satellite triangulation? Or an approach how it can be used here
", "rawMarkdown": "Can you share something for satellite triangulation? Or an approach how it can be used here", "votes": null }, { "id": "1349696", "postDate": "06/15/2021 03:12:48", "content": "Thanks for sharing the steps to start
", "rawMarkdown": "Thanks for sharing the steps to start", "votes": null }, { "id": "1350100", "postDate": "06/15/2021 09:10:04", "content": "Some good resources I found as a novice:
\ntalk from google io (you can skip the first wifi part)
\nhttps://www.gpsworld.com/how-to-achieve-1-meter-accuracy-in-android/
GNSS/INS Integration
\nhttps://ec.europa.eu/research/participants/documents/downloadPublic?documentIds=080166e5c206c883&appId=PPGMS
RTKlibray
\nRTKLIB: An Open Source Program Package for GNSS Positioning
+1 to RTKLIB. Emlid's community forum was great while they sold older GPS modules which needed to be manually post-processed. It's still a good place to go to learn how to use RTKLIB. https://community.emlid.com/c/reach/postprocessing
", "rawMarkdown": "1 to RTKLIB. Emlid's community forum was great while they sold older GPS modules which needed to be manually post-processed. It's still a good place to go to learn how to use RTKLIB. https://community.emlid.com/c/reach/postprocessing", "votes": null }, { "id": "1380676", "postDate": "07/08/2021 08:12:46", "content": "This instructional video helped me in the past understand about the GEOID model which is the gravitationally adjusted elevation model which is different from the GNSS orbit. https://www.youtube.com/watch?v=vOJ3u7Zd_i0
", "rawMarkdown": "This instructional video helped me in the past understand about the GEOID model which is the gravitationally adjusted elevation model which is different from the GNSS orbit. https://www.youtube.com/watch?v=vOJ3u7Zd_i0", "votes": null }, { "id": "1807323", "postDate": "05/31/2022 23:19:18", "content": "LOL - I guess that’s one way to do it
", "rawMarkdown": "LOL - I guess that’s one way to do it", "votes": null } ], "comments": [ { "id": 1304577, "author_name": "ihelon", "author_url": "", "post_date": "05/12/2021 17:51:52", "content": "LOL - I guess that’s one way to do it
", "votes": null, "replies": [] } ] }, { "id": 1304890, "author_name": "emaerthin", "author_url": "", "post_date": "05/13/2021 00:59:18", "content": "thanks for the list @sohier! I started from 1. by applying the Kalman Filter approach naively. One more thought before trying out more advanced ideas: if GPS is used while driving, aligning the signal to the network of roads and preventing from jumping from one street to another street in no time should also improve accuracy.
", "votes": null, "replies": [] }, { "id": 1306938, "author_name": "saurabhbagchi", "author_url": "", "post_date": "05/14/2021 06:42:42", "content": "Thanks for sharing the list @sohier! Ensembling models can also help here is my feeling. Fingers crossed 🙏
", "votes": null, "replies": [] }, { "id": 1322314, "author_name": "makotktjm", "author_url": "", "post_date": "05/25/2021 11:08:55", "content": "What is \"smoothing\" in this context?
\nIt's not \"Label smoothing\" right?
GPS from a phone can be quite \"jittery\" (moves around quite a bit), so some simple smoothing/averaging (especially if the runs have more than one phone that you can combine) can be a pretty good start at getting a reasonable looking path.
", "votes": null, "replies": [] } ] }, { "id": 1331162, "author_name": "tensorchoko", "author_url": "", "post_date": "06/01/2021 10:02:10", "content": "thank you. i think its diffcult for me……
", "votes": null, "replies": [] }, { "id": 1333817, "author_name": "usaiprashanth", "author_url": "", "post_date": "06/03/2021 04:43:11", "content": "Has anyone tried trilateration? I would like to know the results It produces.
\nI've tried it on my own, but I'm getting nan as resultant latitude and longitude after trilateration.
\nI've tried incorporating code from:
\nhttps://gis.stackexchange.com/questions/66/trilateration-using-3-latitude-longitude-points-and-3-distances/415#415
Can you share something for satellite triangulation? Or an approach how it can be used here
", "votes": null, "replies": [] }, { "id": 1349696, "author_name": "manojkrchhetri", "author_url": "", "post_date": "06/15/2021 03:12:48", "content": "Thanks for sharing the steps to start
", "votes": null, "replies": [] }, { "id": 1350100, "author_name": "aemado", "author_url": "", "post_date": "06/15/2021 09:10:04", "content": "Some good resources I found as a novice:
\ntalk from google io (you can skip the first wifi part)
\nhttps://www.gpsworld.com/how-to-achieve-1-meter-accuracy-in-android/
GNSS/INS Integration
\nhttps://ec.europa.eu/research/participants/documents/downloadPublic?documentIds=080166e5c206c883&appId=PPGMS
RTKlibray
\nRTKLIB: An Open Source Program Package for GNSS Positioning
+1 to RTKLIB. Emlid's community forum was great while they sold older GPS modules which needed to be manually post-processed. It's still a good place to go to learn how to use RTKLIB. https://community.emlid.com/c/reach/postprocessing
", "votes": null, "replies": [] } ] }, { "id": 1380676, "author_name": "miniapp", "author_url": "", "post_date": "07/08/2021 08:12:46", "content": "This instructional video helped me in the past understand about the GEOID model which is the gravitationally adjusted elevation model which is different from the GNSS orbit. https://www.youtube.com/watch?v=vOJ3u7Zd_i0
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1304570": "The GPS field presents a lot of domain knowledge that can appear daunting at first. If you're new to the field I would recommend looking at the problem in these tiers, in order of complexity and/or how much domain knowledge is required:\n\n1. Smoothing out the baseline estimates. This requires essentially no domain knowledge.\n2. Integrating readings from other phone instruments, like the accelerometer.\n3. Satellite triangulation using the `*derived.csv` files. \n4. Building triangulations directly from the raw gnss logs. This may require more time to get working, but should also offer more opportunities for error correction.\n5. Incorporating external data for controls like satellite readings from base stations in the area.\n\nThis list shouldn't be considered definitive, and we'd love to hear other ideas about how the machine learning community can get started most efficiently!", "1304577": "1. put all data into a neural network😄", "1304890": "thanks for the list @sohier! I started from 1. by applying the Kalman Filter approach naively. One more thought before trying out more advanced ideas: if GPS is used while driving, aligning the signal to the network of roads and preventing from jumping from one street to another street in no time should also improve accuracy.", "1306938": "Thanks for sharing the list @sohier! Ensembling models can also help here is my feeling. Fingers crossed 🙏", "1322314": "What is \"smoothing\" in this context?\nIt's not \"Label smoothing\" right?", "1322443": "GPS from a phone can be quite \"jittery\" (moves around quite a bit), so some simple smoothing/averaging (especially if the runs have more than one phone that you can combine) can be a pretty good start at getting a reasonable looking path.", "1331162": "thank you. i think its diffcult for me......", "1333817": "Has anyone tried trilateration? I would like to know the results It produces.\n\nI've tried it on my own, but I'm getting nan as resultant latitude and longitude after trilateration.\n\nI've tried incorporating code from:\n[https://gis.stackexchange.com/questions/66/trilateration-using-3-latitude-longitude-points-and-3-distances/415#415](here)", "1345926": "Can you share something for satellite triangulation? Or an approach how it can be used here", "1349696": "Thanks for sharing the steps to start", "1350100": "Some good resources I found as a novice:\n\ntalk from google io (you can skip the first wifi part)\nhttps://www.gpsworld.com/how-to-achieve-1-meter-accuracy-in-android/\n\nGNSS/INS Integration\nhttps://ec.europa.eu/research/participants/documents/downloadPublic?documentIds=080166e5c206c883&appId=PPGMS\n\nRTKlibray\nRTKLIB: An Open Source Program Package for GNSS Positioning", "1380670": "1 to RTKLIB. Emlid's community forum was great while they sold older GPS modules which needed to be manually post-processed. It's still a good place to go to learn how to use RTKLIB. https://community.emlid.com/c/reach/postprocessing", "1380676": "This instructional video helped me in the past understand about the GEOID model which is the gravitationally adjusted elevation model which is different from the GNSS orbit. https://www.youtube.com/watch?v=vOJ3u7Zd_i0", "1807323": "LOL - I guess that’s one way to do it" }, "source": "meta" }