Real-Time Vision Based Human Height Measurement Using Sliding Window on Selected Candidates

This paper presents a real-time human height estimation using image sequences obtained from a single calibrated camera. For each image frame, the candidate value of human height is calculated based on the approximated 3D ground plane and 2D positions of human head top and foot bottom. Some candidates whose values do not differ much from the previous height estimate are selected; a sliding window is then applied on an array of those candidate height values; candidate heights bounded inside the sliding window that has the maximum votes and some candidate heights around that window are finally used in final height estimate. The proposed algorithm of sliding window on selected candidate helps achieve the average accuracy of height estimate at 99.05%; it is a 1.30% increase of accuracy when compared to the final height estimate from all candidates. The standard deviation also decreases from 4.26 to 1.58; the proposed method is superior in terms of accuracy and stability.