A common way to observe animals in nature is to use motion triggered cameras that are called camera-traps. With the expanding usage of camera-trap due to advances in digital technology, the number of images that are collected from camera-traps has increased significantly. Labeling and grouping of animals in these images have put enormous workload on wild-life researchers. We propose a system that frees time for researchers by eliminating useless images-too bright, too dark, too blurred images and images that contain no animals from raw camera-trap data. Firstly, we utilise image histograms to eliminate too bright and too dark images and Fast Fourier Transform to eliminate blurred ones. Secondly, we make use of deep learning techniques and background subtraction to eliminate images without animals and we present the result of our experiments on these subjects. Our approach on eliminating too bright and too dark images have missed very few images and on eliminating blur images we achieve 95.5% success. Finally we show that the technique we propose eliminates more than 50% of images without animals while containing 99% of images with animals.