Segmentasi dan pengesanan objek bergerak dalam keadaan cuaca berjerebu dan berkabus

Abstract

Segmentation and detection of moving object are very important in navigation applications to improve visibility of computer vision technology. The challenges to these issues are how these two issues address hazy and foggy weather. This situation affects technology and specifically the video data used to detect moving objects. This problem occurs due to the light that is scattered because of the fog and haze pixels which prevent light from penetrating resulting in over segmentation. Various methods have been used to improve accuracy and sensitivity in over segmentation but further enhancement is needed to improve the performance in the detection of moving objects. In this research, a new method is proposed to overcome over segmentation which is a combination between Gaussian Mixture Model and other filters based on their own specialities. The combined filters comprised Median Filter and Average Filter for over segmentation, Morphology Filter and Gaussian Filter to rebuild structure element of pixel object, and combination of Blob Analysis, Bounding Box and Kalman Filter to reduce False Positive detection. The combination of these filters is known as Object of Interest Movement (OIM). Qualitative and quantitative methods were used to make comparison with previous methods. Data comprised sources of haze recordings obtained from YouTube and open dataset from Karlsure. Comparative analysis of pictures and calculations of detection of objects were done. Result showed that the combined filters is capable of improving accuracy and sensitivity of the segmentation and detection which were 72.24% for foggy videos, and 76.73% in hazy weather. Based on the findings, the OIM method has proven its capability to improve the accuracy of segmentation and detection object without the need for enhancement to contrast an image.