Reference-Based Human Height Estimation

Human height estimation is one of the research fields used in digital forensics to assist forensic experts in digital image identification. Most of the research was specific to controlled environments, causing hard estimates of many variants of images. In other cases, estimating human height without additional information about the object is sometimes difficult. Therefore, we propose a human height estimation model based on reference objects. The proposed method consists of two significant steps. First, we calculate the height of the person and reference object in pixels. Second, we formulate a linear regression model with the 141 dataset that we calculated before. Then, we evaluate the model with two scenarios, normal and abnormal. The average estimation error is 1.76% with 72 data tests when tested in a typical scenario. Then, we have some results from the abnormal scenarios, i.e., 3.92 % for the arbitrary model, 1.56 % when there is no distance data, and 2.9% when the images are without reference objects. Although our model is not the best in terms of accuracy of estimation results, the error is relatively low considering the method we use and the image variant that can be estimated using our model.