KSC 2020
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
DMoD: When the Sharp Gradient Direction Detects Moving Objects more Effectively than Deep Learning-based Methods |
¿µ¹®Á¦¸ñ(English Title) |
DMoD: When the Sharp Gradient Direction Detects Moving Objects more Effectively than Deep Learning-based Methods |
ÀúÀÚ(Author) |
Md Alamgir Hossain
Md Imtiaz Hossain
Md Delowar Hossain
Ngo Thien Thu
Eui-Nam Huh
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 47 NO. 02 PP. 0951 ~ 0953 (2020. 12) |
Çѱ۳»¿ë (Korean Abstract) |
|
¿µ¹®³»¿ë (English Abstract) |
Moving object detection is the detection of moving targets from an input video. Academy and industry have much interest in moving target detection. In the IoT era, moving object detection plays a key role in object recognition, and tracking for an autonomous car, military, smart home, smart city, health care, agriculture, etc. Additionally, moving object detection is applied for the surveillance of humans, animal, insect, and the environment. However, recent deep learning and traditional shallow learning approaches do not detect moving things more accurately in a complex video including dynamic background, thermal effects, bad weather, pan-tilt-zoom, and turbulence. To address the problem, we propose a moving object detection method that uses more sharp gradient directions and RGB color features. Based on the amount of background change, the decision thresholds are modified and background pixels in the background models are updated dynamically. Additionally, we employ a flood fill based post-processing operation to more effectively filter the raw segmented frame. Our proposed method achieves the best accuracy for most videos of the CDNet-2014 benchmark dataset.
|
Å°¿öµå(Keyword) |
Background Subtraction
Change Detection
Moving Object Detection
Foreground Segmentation
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|