Human Detection Based on Large Feature Sets Using Graphics Processing Units

To obtain a better representation for human detection, edge, color, and texture information have been combined and employed. However, this combination results in an extremely high-dimensional feature space. The large number of feature descriptors results in expensive feature extraction and requires a dimension reduction process. Frameworks based on general purpose graphics processing unit (GPU) programming have been successfully applied in computer vision problems and in this work we model the human detection problem so that multi-core CPUs and multiple GPU devices can be used to speed-up the process. The experimental results show significant reduction on computational time when compared to the serial CPU based approach.