Practical Implementation of A Real-time Human Detection with HOG-AdaBoost in FPGA

Abstract

We reported the practical implementation of a real-time image-based human detection in FPGA. The Histogram of Oriented Gradients (HOG) features and the AdaBoost classifiers are used as an approach. The systolic array architecture based Support Vectoring Machine (SVM) processing is also implemented in our system. According to the results, it can be shown that the humans are successfully detected from a 1280 x 1024 of image resolution with 129 fps of frame rate, it is not only from the front and back views (horizontal axis) but also robust in human detection from different angles (vertical axis). We also compared our architecture with other works.