Higher order symmetry for non-linear classification of human walk detection

The paper focuses on motion-based information extraction from cluttered video image sequences. A novel method is introduced which can reliably detect walking human figures contained in such images. The method works with spatio-temporal input information to detect and classify patterns typical of human movement. Our algorithm consists of real-time operations, which is an important factor in practical applications. The paper presents a new information-extraction and temporal tracking method based on a simplified version of the symmetry-pattern extraction, which pattern is characteristic for the moving legs of a walking person. These spatio-temporal traces are labelled by kernel Fisher discriminant analysis. With the use of temporal tracking and non-linear classification we have achieved pedestrian detection from cluttered image scenes with a correct classification rate of 97.6% from 1 to 2 step periods. The detection rates of linear classifier and SVM are also presented in the results hereby the necessity of a non-linear method and the power of KFDA for this detection task is also demonstrated.