Á¤º¸°úÇÐȸ ³í¹®Áö C : ÄÄÇ»ÆÃÀÇ ½ÇÁ¦
ÇѱÛÁ¦¸ñ(Korean Title) |
´Ù¾çÇÑ »ç¶÷ ¹æÇâÀ» °í·ÁÇÑ ÆÄÆ® ¿µ¿ª ±â¹Ý »ç¶÷ ¿µ¿ª °ËÃâ |
¿µ¹®Á¦¸ñ(English Title) |
Part-based Human Detection Considering Multi-view of Human |
ÀúÀÚ(Author) |
½ÅÁöÇý
À念¹Î
±è»ó¿í
Rammohan Mallipeddi
¹èÁ¤¿Á
ÃÖ¼º¹¬
À̹ÎÈ£
Jihye Shin
Young-Min Jang
Sangwook Kim
Rammohan Mallipeddi
Jungok Bae
Sungmook Choi
Minho Lee
¼ÕÁ¤Àº
Á¤ÁöÈÆ
°íº´Ã¶
³²Àç¿
JungEun Son
JiHun Jung
ByoungChul Ko
JaeYeal Nam
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 19 NO. 11 PP. 0596 ~ 0600 (2013. 11) |
Çѱ۳»¿ë (Korean Abstract) |
º» ³í¹®¿¡¼´Â ÆÄÆ® ¿µ¿ª ±â¹ý°ú ·£´ý Æ÷·¹½ºÆ® ºÐ·ù±â¸¦ ÀÌ¿ëÇÏ¿© Á¤Áö ¿µ»ó¿¡¼ »ç¶÷ Å©±âÀÇ º¯È, ¹è°æÀÇ º¹Àâµµ, ºÎºÐÀû °¡·ÁÁü µî¿¡ °°ÇÇÑ »ç¶÷ °´Ã¼ °ËÃâ ¾Ë°í¸®ÁòÀ» Á¦¾ÈÇÑ´Ù. ¸ÕÀú ¸ñÇ¥ »ç¶÷ °´Ã¼¸¦ Æ÷ÇÔÇÏ´Â À©µµ¿ì¿¡¼ N°³ÀÇ ÆÄÆ® Èĺ¸ ¿µ¿ªÀ» ÀÓÀÇÀÇ Å©±â¿Í À§Ä¡¿¡ »ý¼ºÇÑ´Ù. ÀÌÈÄ ·£´ý Æ÷·¹½ºÆ® ºÐ·ù±â¸¦ ÀÌ¿ëÇÏ¿© »ç¶÷ °ËÃâ¿¡ »ç¿ëÇÒ »óÀ§M°³ÀÇ ÆÄÆ® ¿µ¿ªÀ» »ç¶÷ÀÇ 3°¡Áö ¹æÇâ(Á¤¸é, Ãø¸é, µÞ¸é)¸¶´Ù °áÁ¤ÇÑ´Ù. Å×½ºÆ® ¿µ»ó¿¡ ´ëÇØ Å½»ö À©µµ¿ì¸¦ ¼³Á¤ÇÏ°í, °¢ ¹æÇ⸶´Ù M°³ÀÇ ·ÎÄà ¿µ¿ªÀ» ÇнÀµÈ ·£´ý Æ÷·¹½ºÆ® ºÐ·ù±â¿¡ Àû¿ëÇÑ´Ù. °¢ ¹æÇâ º° ÆÄÆ® ¿µ¿ªÀÇ È®·ü °ªÀ» ÃßÁ¤ÇÏ°í, ÃßÁ¤µÈ È®·ü °ªÀ» °áÇÕÇÏ¿© ÃÖ°í°ªÀ» °®´Â ¹æÇâÀ» ¼±ÅÃÇÑ´Ù. ¼º´É Æò°¡¸¦ À§ÇØ Dalal[1] µî°ú Bourdev[3] µî¿¡¼ Á¦¾ÈÇÑ »ç¶÷ °ËÃâ ¾Ë°í¸®Áò°ú ¼º´É ºñ±³ ½ÇÇèÀ» ÇÏ¿´°í, ±× °á°ú ¼º´É ¸é¿¡¼ ¿ì¼öÇÒ »Ó ¾Æ´Ï¶ó, ÇÁ·Î±×·¥ ½ÇÇà½Ã°£À» »ó´çºÎºÐ ´ÜÃà½ÃÄ×À½À» ½ÇÇè °á°ú¸¦ ÅëÇØ È®ÀÎÇÏ¿´´Ù. |
¿µ¹®³»¿ë (English Abstract) |
In this paper we developed a novel method for detecting multi-view humans in still images that is robust to variation of human size, cluttered background, and partial occlusion using part region and random forest
classifier. To select the appropriate location and size for a part region, we generate a random set of rectangular
part region. Then, we use the random forest classifier to determine M part regions in three viewpoints (front,
profile, back). Given a test image, the selected part features are extracted from a search window and each M
part feature is applied to its corresponding classifier. The total probability of search window is obtained based on
the arithmetic average of each distribution for all part classifiers and we select the view of human as the
maximum probability between three views. The experimental results showed that our algorithm improved the
human detection performance compared with two related methods, Dalal [1] and Bourdev [3], and proposed method
reduced the processing time remarkably. |
Å°¿öµå(Keyword) |
eye tracking
intelligent learning tool
self-learning service
foreign language reading comprehension
»ç¶÷°ËÃâ
´ÙÁß ½ÃÁ¡
ÆÄÆ® ¿µ¿ª
·£´ý Æ÷·¹½ºÆ®
human detection
multi-view
part region
random forest
OCS-LBP
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|