We presented an approach for the visual discrimination of children (3 5 years old) from adults
using stride-based properties of their walking style. Trajectories of marked head and ankle
positions for six children and nine adults were used to compute the relative stride and stride
frequency for each walker at di?erent speeds. The distinction between child and adult for these
features is quite strong and reduces the task of categorization to a linear discrimination
test. Using a trained two-class linear perceptron we were able to achieve a correct classi?-
tion rate of 93 95% for our dataset. Given that only two motion features were used to
characterize and di?erentiate children from adults the result is quite - couraging. The use of
natural modes as a means of visual categorization provides a useful bottom-up framework for the
classi?cation and recognition of humans in motion. References 1. A. Baumberg and D. Hogg.
Learning ?exible models from image sequences. In Proc. European Conf. Comp. Vis. pages 299
308 1994. 2. W. Boda W. Tapp and T. Findley. Biomechanical comparison of treadmill and
overground walking. In Proc. Can. Soc. for Biomech. pages 88 89 1994. 3. C. Bregler and J.
Malik. Tracking people with twists and exponential maps. In Proc. Comp. Vis. and Pattern Rec.
pages 8 15 1998. 4. I. Chang and C. Huang. The model-based human body motion analysis system.
Image and Vision Comp. 18(14):1067 1083 2000. 5. D. Gavrila. Pedestrian detection from a
moving vehicle. In Proc. European Conf.
