{"id":27,"date":"2018-06-02T14:09:50","date_gmt":"2018-06-02T18:09:50","guid":{"rendered":"https:\/\/pressbooks.bccampus.ca\/renegade\/chapter\/chapter-2-1-component-check\/"},"modified":"2018-06-13T14:06:17","modified_gmt":"2018-06-13T18:06:17","slug":"chapter-2-1-component-check","status":"publish","type":"chapter","link":"https:\/\/pressbooks.bccampus.ca\/renegade\/chapter\/chapter-2-1-component-check\/","title":{"raw":"Chapter 2.1 \u2013 Dense Point Cloud","rendered":"Chapter 2.1 \u2013 Dense Point Cloud"},"content":{"raw":"

Overview<\/h1>\r\nThe steps below outline the creation of several 3-D map products starting with a high quality sparse point cloud following optimized photo aligment.\r\n

High Accuracy Alignment<\/h2>\r\nThe low quality photo alignment identified approximately 13,185 key tie points between the overlapping photos. These points form the basis for stitching together all 92 photos together into a seamless mosaic.\r\n\r\nClick\u00a0Show Cameras<\/strong>, rotate the view and\u00a0you should see a result similar to the screenshot below;\r\n\r\n[caption id=\"attachment_105\" align=\"aligncenter\" width=\"1024\"]\"Screenshot Rotated sparse point cloud showing image footprints.[\/caption]\r\n\r\nRecognizing the fact that all cameras exhibit some degree of distortion (e.g. radial, tangential, de-centering, etc.), it's important to run an optimization which minimizes the effects of these distortions on the accuracy and precision of objects in the photos.\r\n\r\nClick\u00a0Optimize Cameras<\/strong>, accept the defaults and click\u00a0OK<\/strong>. From\u00a0Workflow<\/strong>,\u00a0select\u00a0Align Photos...<\/strong> and set the parameters according to the screenshot below;\r\n\r\n[caption id=\"attachment_86\" align=\"aligncenter\" width=\"363\"]\"Screenshot High alignment parameters[\/caption]\r\n\r\nResults of the high quality alignment (which may take about 15min to complete) are shown below with 372,566 key tie points identified. If the alignment is taking too long to complete (more than 20min), you can cancel the process and re-run it using lower quality settings and\/or key\/tie point limits.\r\n\r\n[caption id=\"attachment_110\" align=\"aligncenter\" width=\"1024\"]\"Screenshot Results of high quality photo alignment.[\/caption]\r\n\r\n \r\n
\r\n

Photo Alignment<\/strong><\/h3>\r\nSometimes, large data sets with complex, variable terrain (including trees), will result in misalignment of some photos. If any photos remain unaligned after the initial attempt, several steps can be taken to align them.\r\n
    \r\n \t
  1. Re-run\u00a0Align Photos...<\/strong> using higher\u00a0Key point<\/strong> and\u00a0Tie point<\/strong> limits.<\/li>\r\n \t
  2. Select all the unaligned photos (represented as dots instead of footprints in the model) from the\u00a0Cameras<\/strong> folder under the\u00a0BCIT_Field<\/strong> chunk, right-click, select\u00a0Reset Camera Alignment<\/strong> then right-click again and select\u00a0Align Selected Cameras<\/strong>,<\/li>\r\n \t
  3. Running a higher quality alignment (if possible) can also help align photos.<\/li>\r\n<\/ol>\r\nKeep in mind that poor quality photos (quality index <0.6) or photos with very tall structures like trees or buildings that are significantly closer to the camera than the ground, may not align at all.\r\n\r\n<\/div>\r\nWith the key points identified as accurately as possible, the next step is to locate as many of the remaining pixels in 3-D and generate the dense point cloud.\r\n

    Dense Point Cloud<\/h2>\r\nFrom the\u00a0Workflow<\/strong> menu, select\u00a0Build Dense Cloud...<\/strong>\u00a0The Quality<\/strong>\u00a0setting is directly dependent on available computing power. Reference this\u00a0document<\/a>\u00a0to determine the suggested highest quality level based on the hardware being used. Set the quality as high as possible given your hardware configuration, as this ensures that the maximum number of pixels will be correlated in 3-D space. Be aware that the higher the quality setting, the longer the processing time. Consider using lower setting for large data sets (i.e. several hundred photos). Suggested settings are shown below;\r\n\r\n[caption id=\"attachment_88\" align=\"aligncenter\" width=\"361\"]\"A Dense point cloud parameters[\/caption]\r\n\r\n
    \r\n

    Maximum Density Cloud<\/strong><\/h3>\r\nThe approximate maximum number of points in a dense cloud is calculated by multiplying the number of megapixels<\/strong> of the camera (e.g. 12 megapixel images are 4000 x 3000 pixels thus contain 12 million pixels each) by\u00a0the\u00a0number of images<\/strong> (92 in this data set) which equals 1,104,000,000. It is impossible to correlate all points and thus the actual number of pixels in a point cloud is generally only a fraction of the total possible (0.1 to 5%).\r\n\r\n<\/div>\r\nThe resulting dense point cloud is similar to the key point, low density cloud but it contains exponentially more 3-D information (39,184,432 3-D pixels!). The result should look similar to the screenshot below, but don't be concerned if the number of points doesn't match exactly.\r\n\r\n[caption id=\"attachment_113\" align=\"aligncenter\" width=\"1024\"]\"Screenshot Dense point cloud[\/caption]\r\n\r\nAlthough from certain perspectives, the dense cloud make look like a \"solid\" surface, it's still just a collection of pixels positioned accurately in space based on the GPS coordinate of the center pixel of each image. Due to GPS error and any remaining distortions in the camera and lens, the actual position of any pixel may vary by upto +<\/span>5m but the error usually much lower (+<\/span>0.5m).\r\n\r\nThe next step is to filter any spurious points or outliers which are not representative of the surface being modeled in 3-D.\r\n
    \r\n

    Filtering Dense Cloud Points<\/strong><\/h3>\r\nThe type of surface greatly affects the ability and accuracy of filtering points. For example, the very level and flat playing field (except for the net) means that we shouldn't expect much deviation from the \"ground\" at all. Conversely in heavily treed areas or where there is high frequency, large-scale changes in elevation (e.g. a boulder field), the precision with which outlying points can be effectively filtered declines significantly.\r\n\r\n<\/div>\r\n

    Filter Dense Cloud<\/h2>\r\nThere are a number of different approaches to filtering points. They fall into two broad categories of semi-automated<\/strong> and manual<\/strong>. Here, only the manual method is demonstrated as it is the most selective and best applied to specific areas.\r\n\r\nThe general process involves rotating the dense cloud to identify individual points or groups of points that are not likely to be representative of the surface being modeled (e.g. points floating several tens of meters above the playing field or any points below the \"surface\" of the field). Once points have been identified, the\u00a0Selection<\/strong> (Rectangle, Circle\u00a0<\/strong>or\u00a0Free-form<\/strong>) is used to highlight the suspected outlier and pressing the\u00a0Delete<\/strong> key on the keyboard removes them.\r\n\r\nThis is an iterative process and done well, greatly improves the chances of generating a quality 3-D surface, which is the next step. See below for a sample of outlying points selected using the rectangle selection tool.\r\n\r\n[caption id=\"attachment_114\" align=\"aligncenter\" width=\"1024\"]\"A Selected points to be filtered[\/caption]\r\n

    Adjust Region<\/h2>\r\nIn the above screenshot, you may notice the faint boundaries of a rectangle surrounding the dense cloud. This is the\u00a0Region<\/strong> and defines the outer boundaries of the data set. It appears slightly tilted relative to the points being modeled. This can be adjusted as follows;\r\n
      \r\n \t
    1. From the\u00a0Model<\/strong> menu, open\u00a0Transform Region<\/strong> and select\u00a0Rotate Region<\/strong><\/li>\r\n \t
    2. Using the \"ball\", adjust the angle of the region to match that of the dense point cloud as closely as possible.<\/li>\r\n \t
    3. Use the other\u00a0Transform Region<\/strong> tools from the drop-down menu including\u00a0Move Region<\/strong>,\u00a0Resize Region<\/strong> and\u00a0Reset Region<\/strong>.<\/li>\r\n<\/ol>\r\n[caption id=\"attachment_115\" align=\"aligncenter\" width=\"1024\"]\"Screenshot Adjusted region with dense point cloud[\/caption]\r\n\r\nRefer to the next part for the creation of a 3-D mesh surface and textures.","rendered":"

      Overview<\/h1>\n

      The steps below outline the creation of several 3-D map products starting with a high quality sparse point cloud following optimized photo aligment.<\/p>\n

      High Accuracy Alignment<\/h2>\n

      The low quality photo alignment identified approximately 13,185 key tie points between the overlapping photos. These points form the basis for stitching together all 92 photos together into a seamless mosaic.<\/p>\n

      Click\u00a0Show Cameras<\/strong>, rotate the view and\u00a0you should see a result similar to the screenshot below;<\/p>\n

      \"Screenshot
      Rotated sparse point cloud showing image footprints.<\/figcaption><\/figure>\n

      Recognizing the fact that all cameras exhibit some degree of distortion (e.g. radial, tangential, de-centering, etc.), it’s important to run an optimization which minimizes the effects of these distortions on the accuracy and precision of objects in the photos.<\/p>\n

      Click\u00a0Optimize Cameras<\/strong>, accept the defaults and click\u00a0OK<\/strong>. From\u00a0Workflow<\/strong>,\u00a0select\u00a0Align Photos…<\/strong> and set the parameters according to the screenshot below;<\/p>\n

      \"Screenshot
      High alignment parameters<\/figcaption><\/figure>\n

      Results of the high quality alignment (which may take about 15min to complete) are shown below with 372,566 key tie points identified. If the alignment is taking too long to complete (more than 20min), you can cancel the process and re-run it using lower quality settings and\/or key\/tie point limits.<\/p>\n

      \"Screenshot
      Results of high quality photo alignment.<\/figcaption><\/figure>\n

       <\/p>\n

      \n

      Photo Alignment<\/strong><\/h3>\n

      Sometimes, large data sets with complex, variable terrain (including trees), will result in misalignment of some photos. If any photos remain unaligned after the initial attempt, several steps can be taken to align them.<\/p>\n

        \n
      1. Re-run\u00a0Align Photos…<\/strong> using higher\u00a0Key point<\/strong> and\u00a0Tie point<\/strong> limits.<\/li>\n
      2. Select all the unaligned photos (represented as dots instead of footprints in the model) from the\u00a0Cameras<\/strong> folder under the\u00a0BCIT_Field<\/strong> chunk, right-click, select\u00a0Reset Camera Alignment<\/strong> then right-click again and select\u00a0Align Selected Cameras<\/strong>,<\/li>\n
      3. Running a higher quality alignment (if possible) can also help align photos.<\/li>\n<\/ol>\n

        Keep in mind that poor quality photos (quality index <0.6) or photos with very tall structures like trees or buildings that are significantly closer to the camera than the ground, may not align at all.<\/p>\n<\/div>\n

        With the key points identified as accurately as possible, the next step is to locate as many of the remaining pixels in 3-D and generate the dense point cloud.<\/p>\n

        Dense Point Cloud<\/h2>\n

        From the\u00a0Workflow<\/strong> menu, select\u00a0Build Dense Cloud…<\/strong>\u00a0The Quality<\/strong>\u00a0setting is directly dependent on available computing power. Reference this\u00a0document<\/a>\u00a0to determine the suggested highest quality level based on the hardware being used. Set the quality as high as possible given your hardware configuration, as this ensures that the maximum number of pixels will be correlated in 3-D space. Be aware that the higher the quality setting, the longer the processing time. Consider using lower setting for large data sets (i.e. several hundred photos). Suggested settings are shown below;<\/p>\n

        \"A
        Dense point cloud parameters<\/figcaption><\/figure>\n
        \n

        Maximum Density Cloud<\/strong><\/h3>\n

        The approximate maximum number of points in a dense cloud is calculated by multiplying the number of megapixels<\/strong> of the camera (e.g. 12 megapixel images are 4000 x 3000 pixels thus contain 12 million pixels each) by\u00a0the\u00a0number of images<\/strong> (92 in this data set) which equals 1,104,000,000. It is impossible to correlate all points and thus the actual number of pixels in a point cloud is generally only a fraction of the total possible (0.1 to 5%).<\/p>\n<\/div>\n

        The resulting dense point cloud is similar to the key point, low density cloud but it contains exponentially more 3-D information (39,184,432 3-D pixels!). The result should look similar to the screenshot below, but don’t be concerned if the number of points doesn’t match exactly.<\/p>\n

        \"Screenshot
        Dense point cloud<\/figcaption><\/figure>\n

        Although from certain perspectives, the dense cloud make look like a “solid” surface, it’s still just a collection of pixels positioned accurately in space based on the GPS coordinate of the center pixel of each image. Due to GPS error and any remaining distortions in the camera and lens, the actual position of any pixel may vary by upto +<\/span>5m but the error usually much lower (+<\/span>0.5m).<\/p>\n

        The next step is to filter any spurious points or outliers which are not representative of the surface being modeled in 3-D.<\/p>\n

        \n

        Filtering Dense Cloud Points<\/strong><\/h3>\n

        The type of surface greatly affects the ability and accuracy of filtering points. For example, the very level and flat playing field (except for the net) means that we shouldn’t expect much deviation from the “ground” at all. Conversely in heavily treed areas or where there is high frequency, large-scale changes in elevation (e.g. a boulder field), the precision with which outlying points can be effectively filtered declines significantly.<\/p>\n<\/div>\n

        Filter Dense Cloud<\/h2>\n

        There are a number of different approaches to filtering points. They fall into two broad categories of semi-automated<\/strong> and manual<\/strong>. Here, only the manual method is demonstrated as it is the most selective and best applied to specific areas.<\/p>\n

        The general process involves rotating the dense cloud to identify individual points or groups of points that are not likely to be representative of the surface being modeled (e.g. points floating several tens of meters above the playing field or any points below the “surface” of the field). Once points have been identified, the\u00a0Selection<\/strong> (Rectangle, Circle\u00a0<\/strong>or\u00a0Free-form<\/strong>) is used to highlight the suspected outlier and pressing the\u00a0Delete<\/strong> key on the keyboard removes them.<\/p>\n

        This is an iterative process and done well, greatly improves the chances of generating a quality 3-D surface, which is the next step. See below for a sample of outlying points selected using the rectangle selection tool.<\/p>\n

        \"A
        Selected points to be filtered<\/figcaption><\/figure>\n

        Adjust Region<\/h2>\n

        In the above screenshot, you may notice the faint boundaries of a rectangle surrounding the dense cloud. This is the\u00a0Region<\/strong> and defines the outer boundaries of the data set. It appears slightly tilted relative to the points being modeled. This can be adjusted as follows;<\/p>\n

          \n
        1. From the\u00a0Model<\/strong> menu, open\u00a0Transform Region<\/strong> and select\u00a0Rotate Region<\/strong><\/li>\n
        2. Using the “ball”, adjust the angle of the region to match that of the dense point cloud as closely as possible.<\/li>\n
        3. Use the other\u00a0Transform Region<\/strong> tools from the drop-down menu including\u00a0Move Region<\/strong>,\u00a0Resize Region<\/strong> and\u00a0Reset Region<\/strong>.<\/li>\n<\/ol>\n
          \"Screenshot
          Adjusted region with dense point cloud<\/figcaption><\/figure>\n

          Refer to the next part for the creation of a 3-D mesh surface and textures.<\/p>\n","protected":false},"author":365,"menu_order":1,"template":"","meta":{"pb_show_title":"on","pb_short_title":"","pb_subtitle":"","pb_authors":[],"pb_section_license":""},"chapter-type":[],"contributor":[],"license":[],"class_list":["post-27","chapter","type-chapter","status-publish","hentry"],"part":19,"_links":{"self":[{"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/pressbooks\/v2\/chapters\/27","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/pressbooks\/v2\/chapters"}],"about":[{"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/wp\/v2\/types\/chapter"}],"author":[{"embeddable":true,"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/wp\/v2\/users\/365"}],"version-history":[{"count":20,"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/pressbooks\/v2\/chapters\/27\/revisions"}],"predecessor-version":[{"id":203,"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/pressbooks\/v2\/chapters\/27\/revisions\/203"}],"part":[{"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/pressbooks\/v2\/parts\/19"}],"metadata":[{"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/pressbooks\/v2\/chapters\/27\/metadata\/"}],"wp:attachment":[{"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/wp\/v2\/media?parent=27"}],"wp:term":[{"taxonomy":"chapter-type","embeddable":true,"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/pressbooks\/v2\/chapter-type?post=27"},{"taxonomy":"contributor","embeddable":true,"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/wp\/v2\/contributor?post=27"},{"taxonomy":"license","embeddable":true,"href":"https:\/\/pressbooks.bccampus.ca\/renegade\/wp-json\/wp\/v2\/license?post=27"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}