With respect to road crash statistics, on-board pedestrian detection is a key task for future advanced driver assistance systems. In this paper, the authors describe the implementation of a real-time pedestrian recognition system that combines FPGA-based extraction of image features with a CPU-based object localization and classification framework. In terms of features, they have implemented the Histograms of Oriented Gradients (HOG) descriptor that is state-of-the-art in the field of human detection from a moving camera. While past HOG-related publications presented simplified FPGA-based HOG variants, often sacrificing classification performance, they implemented the original descriptor with minor modifications on dedicated hardware. Evaluation on the INRIA pedestrian database shows potential for deploying the system in practice. The descriptor computation runs on a PCIe frame grabber with embedded FPGA that can be directly integrated into an automotive computer of a test vehicle for evaluation purposes. The paper is organized as follows: They briefly discuss previous work in section 2. The description of the method in section 3 is followed by implementation details in section 4. Experimental results are presented in section 5, before they conclude with a summary and discuss open issues in section 6.
FPGA implementation of a HOG-based pedestrian recognition system
2009
10 Seiten, 9 Bilder, 2 Tabellen, 30 Quellen
Aufsatz (Konferenz)
Englisch
PEDESTRIAN RECOGNITION DEVICE, PEDESTRIAN RECOGNITION METHOD, AND PROGRAM
Europäisches Patentamt | 2019
|Motion Based Pedestrian Recognition
IEEE | 2008
|CROSSWALK PEDESTRIAN SIGNAL ASSIST SYSTEM WITH IMAGE-BASED PEDESTRIAN RECOGNITION ALGORITHM
Europäisches Patentamt | 2021
|Europäisches Patentamt | 2020
|