ANIMAL DETECTION USING A SERIES OF IMAGES UNDER COMPLEX SHOOTING CONDITIONS

Abstract. Camera traps providing enormous number of images during a season help to observe remotely animals in the wild. However, analysis of such image collection manually is impossible. In this research, we develop a method for automatic animal detection based on background modeling of scene under complex shooting. First, we design a fast algorithm for image selection without motions. Second, the images are processed by modified Multi-Scale Retinex algorithm in order to align uneven illumination. Finally, background is subtracted from incoming image using adaptive threshold. A threshold value is adjusted by saliency map, which is calculated using pyramid consisting of the original image and images modified by MSR algorithm. Proposed method allows to achieve high estimators of animals detection.