¼ö·Ï»çÇ× |
Çѱ¹±¸Á¶¹°Áø´ÜÀ¯Áö°ü¸®°øÇÐȸ ³í¹®Áý , Vol.29 No.2(2025-04) |
ÁÖÁ¦¾î |
ÀΰøÁö´É; ºñºÎÂø; ºñÁ¢ÃË; YOLO; °èÃø¹æ¹ý ; AI; Non-attachment; Non-contact; YOLO; Measurement method |
¿ä¾à1 |
¹®ÈÀç µî ÁÖ¿ä ½Ã¼³¹°ÀÇ º¯À§ °èÃø¿¡´Â ´ë»ó½Ã¼³ÀÇ Ç¥¸é¿¡ ºÎÂøÇÏ´Â ±âÁ¸ ¹æ½ÄÀÇ °èÃøÀÌ ¾Æ´Ñ ºñºÎÂø, ºñÁ¢ÃË °³³äÀÇ ½Å±â¼úÀÌ ¿ä±¸µÇ°í ÀÖ´Ù. º» ¿¬±¸¿¡¼´Â 1m °Å¸®¿¡¼ 0.1§®¹Ì¸¸ÀÇ ºÐÇØ´ÉÀ» °¡Áø È»óºÐ¼® ±â¹Ý °èÃø ¾Ë°í¸®ÁòÀ» °³¹ßÇÏ¿´´Ù. ·¹ÀÌÀú¸¦ ÅëÇÑ °¡»ó ±×¸®µå¸¦ Çü¼ºÇÏ°í ¿µ»óÀ» ÅëÇØ ¸ð´ÏÅ͸µÇÏ´Â Àåºñ¸¦ °í¾ÈÇÏ¿´À¸¸ç, ÁÖ¿ä ½Ã¼³¹°¿¡ ¼³Ä¡ÇÏ°í ¿î¿ëµÇ´Â 4³â µ¿¾ÈÀÇ °úÁ¤¿¡¼ ¼öÁýµÈ °¢Á¾ µ¥ÀÌÅ͸¦È°¿ëÇÏ´Â ¹æ¾ÈÀ» °í·ÁÇÏ¿´´Ù. ÇöÀå ¿î¿ëÁßÀÎ °èÃø±â´Â ½Ç³»½ÃÇèÀ¸·Î ºÐÇØ´É ¹× Á¤È®µµ¸¦ °ËÁõÇÏ¿´À¸³ª, ÇöÀå Ȱ¿ë¿¡ ÀÖ¾î ÁÖ?¾ß°£ÀÇ Á¶µµÂ÷¹× °¿ì, ¾È°³ µî ±â»ó ¾ÇÁ¶°ÇÀ¸·Î ÀÎÇÑ ³ëÀÌÁî ¹®Á¦ÀÇ °³¼±ÀÌ ÇÊ¿äÇÏ¿´´Ù. ÀÌ¿¡ AI ¾Ë°í¸®ÁòÀ» Ȱ¿ëÇÏ¿© À̹ÌÁö ÆÇµ¶ÀÇ Á¤È®µµ¸¦ È®º¸ÇϰíÀÚÇÏ¿´´Ù. Àû¿ëµÈ °¡»ó ±×¸®µå ÆÐÅÏÀÇ ÀÚµ¿ÀνÄÀ» À§ÇÏ¿© YOLO ¾Ë°í¸®ÁòÀ» Àû¿ëÇÏ¿´À¸¸ç, ÇнÀµ¥ÀÌÅÍ·Î ÇöÀå¿¡¼ ȹµæÇÑ Á¤»ó ¹× ³ëÀÌÁî Æ÷ÇÔÀ̹ÌÁö ÃÑ 6,512 ÀåÀ» »ç¿ëÇÏ¿´°í IOU´Â 0.9·Î È®º¸ÇÏ¿´´Ù. ÇнÀ°á°ú ÀÏÆò±Õ µ¥ÀÌÅÍÀÇ ÆíÂ÷°¡ ±âÁ¸ ¡¾40Çȼ¿(1.6mm)¿¡¼ °³¹ß ¾Ë°í¸®Áò Àû¿ëÈÄ ¡¾20Çȼ¿(0.8mm)À¸·Î °³¼±µÇ¾î AI ¾Ë°í¸®ÁòÀ» ÅëÇÑ ³ëÀÌÁî °³¼± È¿°ú¸¦ ÀÔÁõÇÏ¿´´Ù. |
¿ä¾à2 |
For displacement measurement of cultural properties and important structures, new technologies based on non-contact, non-attachment methods are required instead of traditional contact-based measurements. This study developed an image-based measurement algorithm with a resolution of less than 0.1 mm at a 1-meter distance. A laser-generated virtual grid system was designed to monitor structures through imaging, and the system was tested and operated over four years of data collection at various facilities. The measurement device was verified for resolution and accuracy through indoor testing, but field deployment faced challenges such as noise due to varying light conditions between day and night, as well as adverse weather conditions like rain and fog. To enhance I mage reading accuracy, an AI algorithm was applied. Specifically, the YOLO algorithm was utilized for automatic recognition of the virtual grid pattern, using a dataset consisting of a total of 6,512 images from field conditions, including normal and noisy samples, achieving an Intersection over Union (IoU) score of 0.9. The application of the developed algorithm demonstrated improved daily data variance, reducing from ¡¾40 pixels (1.6mm) to ¡¾20 pixels (0.8mm), thereby validating the noise reduction effectiveness through AI integration. |