{
  "id": 1633,
  "title": "Segmentation of a video",
  "url": "/competitions/GestureChallenge/discussion/1633",
  "author_name": "",
  "post_date": "2012-03-25T16:01:09.703Z",
  "votes": null,
  "comment_count": 1,
  "views": 2246,
  "content": "<p>What is the simplest way of segmenting a video</p>\r\n<p>I mean how can we find start and end a gesture in a test video ?</p>\r\n<p>looking forward for your cooperation</p>\r\n<p>thanks</p>",
  "messages": [
    {
      "id": "9735",
      "postDate": "03/25/2012 16:01:09",
      "content": "<p>What is the simplest way of segmenting a video</p>\r\n<p>I mean how can we find start and end a gesture in a test video ?</p>\r\n<p>looking forward for your cooperation</p>\r\n<p>thanks</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "9742",
      "postDate": "03/26/2012 03:41:43",
      "content": "<p>[quote=synopsis360;9735]</p>\r\n<p>What is the simplest way of segmenting a video</p>\r\n<p>[/quote]</p>\r\n<p><br>\r\nHere's a simple algorithm that I have been using:<br>\r\n<br>\r\n- Compute the correlations of all frames with the first frame.<br>\r\n<br>\r\n- If there is a run of frames with high correlations (say, more than one standard deviation away from the mean correlation), assume that it is a boundary.<br>\r\n<br>\r\nThe rationale is that the gesture boundaries would look similar to the starting position. I am still trying to figure out the best way to measure the correlation between frames. Inverse Euclidean distance or Pearson should work.<br>\r\n<br>\r\nYou could get fancy and take the correlation with the surrounding frames into account. This exploits the assumption that the user pauses for some time between individual gestures. Correlation with the last frame would be another useful input.<br>\r\n<br>\r\nYou should note that I am not very high on the Leaderboard, though.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 9742,
      "author_name": "anlthms",
      "author_url": "",
      "post_date": "03/26/2012 03:41:43",
      "content": "<p>[quote=synopsis360;9735]</p>\r\n<p>What is the simplest way of segmenting a video</p>\r\n<p>[/quote]</p>\r\n<p><br>\r\nHere's a simple algorithm that I have been using:<br>\r\n<br>\r\n- Compute the correlations of all frames with the first frame.<br>\r\n<br>\r\n- If there is a run of frames with high correlations (say, more than one standard deviation away from the mean correlation), assume that it is a boundary.<br>\r\n<br>\r\nThe rationale is that the gesture boundaries would look similar to the starting position. I am still trying to figure out the best way to measure the correlation between frames. Inverse Euclidean distance or Pearson should work.<br>\r\n<br>\r\nYou could get fancy and take the correlation with the surrounding frames into account. This exploits the assumption that the user pauses for some time between individual gestures. Correlation with the last frame would be another useful input.<br>\r\n<br>\r\nYou should note that I am not very high on the Leaderboard, though.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "9735": "",
    "9742": ""
  },
  "source": "meta"
}