A Multi-Type Features Method for Leg Detection in 2-D Laser Range Data

People detection is an important topic in the fields of security, intelligent environments, and robotics. Current research on people detection based on a single laser range finder is mostly focused on leg detection. However, in practical environments, where legs are likely to be touching or partially occluded, the current methods suffer from a low detection rate and precision. This paper proposes a multi-type features method for leg detection in 2-D laser range data. This method consists of segmentation, through which the laser range data are divided into segments; feature definition and extraction, in which three types of features, including relative distance statistical features, spatial relationship features and nearest neighbor features, are introduced and combined with classic geometric features; and classification, by which a strong classifier is generated using the real AdaBoost algorithm and segments are classified as leg or non-leg. Three 2-D laser range data sets are used for the experiments. The experimental results show that the proposed features are robust and effective in detecting both separated legs and touched or partially occluded legs.