An Encoder-Decoder Convolutional Neural Network for Change Detection

Abstract

Recently, Convolutional Neural Network has shown great performance for many computer vision tasks, including change detection. We adopt the idea of encoder-decoder structured convolutional neural network for background subtraction and foreground object segmentation. In our CNN-based background subtraction system, deep features from target frame and reference frame are extracted and compared to estimate the difference in encoder part. Then the decoder converts these features from encoder to into segmentation map with fine detail. The experimental results tested on CDNet 2014 dataset show that the proposed structure archives the state-of-the-art performance.