Forensic photo/videogrammetry; Monte Carlo simulation of pixel and measurement errors

In this paper, we present some results from a study in progress on methods for the measurement of the length of a robber in a surveillance video image. A calibration tool was constructed for the calibration of the camera. Standard procedures for computing the lens distortion, image projection parameters and the length of the robber have been implemented in Mathematika. These procedures are based on the use of pixel coordinates of markers on the calibration tool, the robber's head (and, optionally, his feet) and an estimation of his position in the coordinate system that is defined by the calibration tool. Monte-Carlo simulation is used to compute a histogram of the robber's length, yielding an estimation of minimum and maximum values. In a repeated process pixel and position coordinates are selected randomly from predefined ranges and, using these data sets, the length, position and quality of the fit are computed and stored. The range of the length in the histogram can be made smaller by selecting only those data sets that can meet one or more constraints, e.g. the quality of the fit should be good and the position of the robber should be within the physical limits of the scene. Some experimental results are presented and discussed.