FPGA Implementation of HOG-based Real-Time Object Detection Processor

摘要

Histogram of Oriented Gradients (HOG) is widely accepted feature descriptor for object detection. HOG is robust against changes in illumination and texture. Thus, HOG is highly effective for pedestrian detection and other object detection. In recent years, object detection techniques have been introduced to advanced automobile products and increasingly valuable. Additionally, recent progress of general-purpose processors enables to implement algorithms that require heavy computations such as HOG-based object detection. However, these processors suffer from nigh power consumption and are therefore unsuitable for mobile systems under limited battery conditions. Therefore, we propose a HOG-based object detection calculation, cell-based pipeline architecture and parallelized modules. To evaluate the effectiveness of our approach, the proposed architecture is implemented onto a FPGA prototyping board. Results show that the proposed architecture can generate HOG features and detect objects with 40 MHz for SVGA resolution video (800×600 pixels) at 72 frames per second (fps).