Pedestrian speed estimation based on direct linear transformation calibration

The speed of pedestrian is an important factor in behavior analysis of pedestrian. Speed calculation of pedestrian can be equivalent to the analysis of the motion trails of object. It can be analyzed in two parts. The first part is the coordinate transformation between the world coordinate system and the image coordinate system which is accomplished by “Distance-to-Pixel” mapping table. The second part is the compute of real motion distance. In this paper, the mapping table is achieved by calibrating camera using DLT (Direct Linear Transformation) method. After that, the motion trajectory in 3D space can be obtained by transforming the motion trajectory in image and be divided into several parts so that every part is approximately linear. The estimated final speed is the mean value of the speeds in every linear part. Experimental results have demonstrated that the mapping table can measure distances within 2%. Our process works in real time and produces speed with over 95% accuracy.