![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F303430%2F7311c0284b923a25a9da59ee32a61848%2Finbox_6537187_701336f11a01e438007d7e6de2989b20_V_ML_AP_Ex.png?generation=1683894691248673&alt=media\")
The usual situation in data series looks like this:
\n
Where AccV is always negative (mean=-1, max_value=-0.2). According to the laws of physics, by the end of the serie (usually 20-30 minutes), the object must have a space velocity: -1 * 1200 = -1200 units / sec. So the question is:
\nWhy is AccV always negative? Bad sensor or what?
", "messages": [ { "id": "2256379", "postDate": "05/12/2023 12:49:31", "content": "The usual situation in data series looks like this:
\n
Where AccV is always negative (mean=-1, max_value=-0.2). According to the laws of physics, by the end of the serie (usually 20-30 minutes), the object must have a space velocity: -1 * 1200 = -1200 units / sec. So the question is:
\nWhy is AccV always negative? Bad sensor or what?
", "rawMarkdown": "The usual situation in data series looks like this:\n\n\n\n\nWhere AccV is always negative (mean=-1, max_value=-0.2). According to the laws of physics, by the end of the serie (usually 20-30 minutes), the object must have a space velocity: -1 * 1200 = -1200 units / sec. So the question is:\n\nWhy is AccV always negative? Bad sensor or what?", "votes": null }, { "id": "2256408", "postDate": "05/12/2023 13:27:48", "content": "I think gravity has an effect.
", "rawMarkdown": "I think gravity has an effect.", "votes": null }, { "id": "2256528", "postDate": "05/12/2023 14:50:39", "content": "I also think about this. but I still don't understand it. It seems that the acceleration of free fall should not affect in any way (since the body remains motionless and is not accelerated). Even there is some kind of influence, then it takes place to be on all measurements, then it turns out that all measurements must be shifted by some constant. Then the question is what value (and it doesn't look like there's a common constant)?
", "rawMarkdown": "I also think about this. but I still don't understand it. It seems that the acceleration of free fall should not affect in any way (since the body remains motionless and is not accelerated). Even there is some kind of influence, then it takes place to be on all measurements, then it turns out that all measurements must be shifted by some constant. Then the question is what value (and it doesn't look like there's a common constant)?", "votes": null }, { "id": "2256546", "postDate": "05/12/2023 15:06:08", "content": "Hey Alexander,
\nAccelerometers like the kind in this competition are measuring all of the accelerations on the device relative to some rest frame, including the Earth's gravity downward and the actual accelerations of the person's motion. The wiki does an okay-ish job explaining: https://en.wikipedia.org/wiki/Accelerometer#Physical_principles
\nThere are various types of accelerometers, but as an example suppose there is a little spring attached to a mass inside this accelerometer. The spring will stretch/compress when the person accelerates. However, it will also stretch just by being pulled by gravity. The physics term here is the Equivalence Principle.
\nIntegrating accelerometer data does not quite work as you have noted. You would need a dead reckoning method (gyroscope + accelerometer, e.g.) to do this. (see this comment)
", "rawMarkdown": "Hey Alexander,\n\nAccelerometers like the kind in this competition are measuring all of the accelerations on the device relative to some rest frame, including the Earth's gravity downward and the actual accelerations of the person's motion. The wiki does an okay-ish job explaining: https://en.wikipedia.org/wiki/Accelerometer#Physical_principles \n\nThere are various types of accelerometers, but as an example suppose there is a little spring attached to a mass inside this accelerometer. The spring will stretch/compress when the person accelerates. However, it will also stretch just by being pulled by gravity. The physics term here is the [Equivalence Principle](https://en.wikipedia.org/wiki/Equivalence_principle). \n\nIntegrating accelerometer data does not quite work as you have noted. You would need a [dead reckoning](https://en.wikipedia.org/wiki/Dead_reckoning) method (gyroscope + accelerometer, e.g.) to do this. (see [this comment](https://www.kaggle.com/competitions/tlvmc-parkinsons-freezing-gait-prediction/discussion/396926#2206797))", "votes": null }, { "id": "2257640", "postDate": "05/13/2023 14:46:06", "content": "It's 1g: exactly gravity…
", "rawMarkdown": "It's 1g: exactly gravity...", "votes": null }, { "id": "2261963", "postDate": "05/16/2023 16:35:29", "content": "Hey, Austin,
\nThanks, a lot. Now I understand why there is always negative value and why it was not shifted by one constant.
", "rawMarkdown": "Hey, Austin,\n\nThanks, a lot. Now I understand why there is always negative value and why it was not shifted by one constant.", "votes": null } ], "comments": [ { "id": 2256408, "author_name": "yutoshimomura", "author_url": "", "post_date": "05/12/2023 13:27:48", "content": "I think gravity has an effect.
", "votes": null, "replies": [ { "id": 2256528, "author_name": "aas1000", "author_url": "", "post_date": "05/12/2023 14:50:39", "content": "I also think about this. but I still don't understand it. It seems that the acceleration of free fall should not affect in any way (since the body remains motionless and is not accelerated). Even there is some kind of influence, then it takes place to be on all measurements, then it turns out that all measurements must be shifted by some constant. Then the question is what value (and it doesn't look like there's a common constant)?
", "votes": null, "replies": [ { "id": 2257640, "author_name": "albertoannoni", "author_url": "", "post_date": "05/13/2023 14:46:06", "content": "It's 1g: exactly gravity…
", "votes": null, "replies": [] } ] } ] }, { "id": 2256546, "author_name": "austinhinkel", "author_url": "", "post_date": "05/12/2023 15:06:08", "content": "Hey Alexander,
\nAccelerometers like the kind in this competition are measuring all of the accelerations on the device relative to some rest frame, including the Earth's gravity downward and the actual accelerations of the person's motion. The wiki does an okay-ish job explaining: https://en.wikipedia.org/wiki/Accelerometer#Physical_principles
\nThere are various types of accelerometers, but as an example suppose there is a little spring attached to a mass inside this accelerometer. The spring will stretch/compress when the person accelerates. However, it will also stretch just by being pulled by gravity. The physics term here is the Equivalence Principle.
\nIntegrating accelerometer data does not quite work as you have noted. You would need a dead reckoning method (gyroscope + accelerometer, e.g.) to do this. (see this comment)
", "votes": null, "replies": [ { "id": 2261963, "author_name": "aas1000", "author_url": "", "post_date": "05/16/2023 16:35:29", "content": "Hey, Austin,
\nThanks, a lot. Now I understand why there is always negative value and why it was not shifted by one constant.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2256379": "The usual situation in data series looks like this:\n\n\n\n\nWhere AccV is always negative (mean=-1, max_value=-0.2). According to the laws of physics, by the end of the serie (usually 20-30 minutes), the object must have a space velocity: -1 * 1200 = -1200 units / sec. So the question is:\n\nWhy is AccV always negative? Bad sensor or what?", "2256408": "I think gravity has an effect.", "2256528": "I also think about this. but I still don't understand it. It seems that the acceleration of free fall should not affect in any way (since the body remains motionless and is not accelerated). Even there is some kind of influence, then it takes place to be on all measurements, then it turns out that all measurements must be shifted by some constant. Then the question is what value (and it doesn't look like there's a common constant)?", "2256546": "Hey Alexander,\n\nAccelerometers like the kind in this competition are measuring all of the accelerations on the device relative to some rest frame, including the Earth's gravity downward and the actual accelerations of the person's motion. The wiki does an okay-ish job explaining: https://en.wikipedia.org/wiki/Accelerometer#Physical_principles \n\nThere are various types of accelerometers, but as an example suppose there is a little spring attached to a mass inside this accelerometer. The spring will stretch/compress when the person accelerates. However, it will also stretch just by being pulled by gravity. The physics term here is the [Equivalence Principle](https://en.wikipedia.org/wiki/Equivalence_principle). \n\nIntegrating accelerometer data does not quite work as you have noted. You would need a [dead reckoning](https://en.wikipedia.org/wiki/Dead_reckoning) method (gyroscope + accelerometer, e.g.) to do this. (see [this comment](https://www.kaggle.com/competitions/tlvmc-parkinsons-freezing-gait-prediction/discussion/396926#2206797))", "2257640": "It's 1g: exactly gravity...", "2261963": "Hey, Austin,\n\nThanks, a lot. Now I understand why there is always negative value and why it was not shifted by one constant." }, "source": "meta" }