Improvised Filter Design for Depth Estimation from Single Monocular Images

Abstract

Depth Estimation poses various challenges and has wide range applications. Techniques for depth prediction from static images include binocular vision, focus-defocus, stereo vision and single monocular images unfortunately not much attention has been paid on depth estimation from single image except [1]. We have proposed a method for depth estimation from single monocular images which is based on filters that are used to extract key image features. The filters used have been applied at multiple scales to take into account local and global features. Here an attempt is made to reduce the dimension of feature vector as proposed in [1]. In this paper we have optimized the filters used for texture gradient extraction. This paper also introduces a prediction algorithm whose parameters are learned by repeated correction. The new methodology proposed provides an equivalent quality of result as in [1].