This paper focuses on distinguishing obstacles from free regions towards navigation of the visually challenged. This is an important and challenging research problem due to the unconstrained navigating environment containing diverse obstacles. It is necessary to detect if any obstacles are lying ahead in advance with high precision which will result in a warning to the person. We first implement Ulrich’s classical obstacle detection method and find its detection accuracy to be high.
However, the higher false positive rate for free path detection motivates us to improve the method. Instead of just comparing reference area with region of interest, we embed the saliency information after adaptive computation of histogram-based saliency threshold. The weightage of initial obstacle map with thresholded saliency information results in a much accurate and desirable obstacle map. The experimental results with a newly created database of unconstrained indoor and outdoor images support the claims made in the paper.