A Study Of Aurora Image Sequence Event Detection

Colorful aurora, which is so mysterious, beautiful and diverse, often occurs in the north and south polar regions. For thousands of years, aurora phenomenon has been attracting human to explore and pursue.Researchers have been using various means to conduct joint observations, using a variety of data such as radar data, satellite data to study the physical change process of aurora in order to investigate the corresponding physical laws. Among them, the aurora images captured by ground equipment are indispensable. Currently, the number of aurora images has reached millions and millions, we need to use the computer to analysis and process the aurora data automatically. In this paper, an in-depth study of automated analysis of aurora image sequence processing using computer vision technology is made.First of all, this paper analyzed the poleward moving arc aurora image sequences. A method named PVLBP(poleward volume local binary patterns), which is based on the existing method VLBP(volume local binary patterns), is proposed. PVLBP only consider the image sequence information in poleward direction and ignore other directions, it’s in line with the motion characteristics of aurora image sequence to be detected. Experimental results show the proposed method can maintain a relatively high detection accuracy and reduce the feature dimension at the same time, improve the detection efficiency.Secondly, aiming at the fact that PVLBP ignore the spatial position information of the single frame image, a structure combining the interframe continuity information of the sequence and the spatial location information of the single frame named spatial-temporal structure is proposed, and a new method called ST-PVLBP(spatial-temporal poleward volume local binary patterns) is proposed based on PVLBP and spatial-temporal structure. The impact on the layers of the spatial-temporal structure is discussed. Experimental results show ST-PVLBP can maintain a high detection accuracy while maintaining high detection efficiency, achieve both accuracy and efficiency.Finally, other types of aurora image sequence event are detected, but not limited to a single target event. The universality of spatial-temporal structure is verified, and the static LBP laws are extended to the dynamic VLBP mode. Then the principle component analysis is applied to VLBP, proposed a method named ST-VLBP-PCA. Experimental results show that the proposed method can be used as an effective mean to describe aurora image sequence feature.