Stereo matching constitutes a critical technology in applications such as autonomous driving and robot navigation. Conventional algorithms, however, often encounter limitations in real-time performance and resource efficiency when deployed on embedded platforms. This paper presents a real-time stereo matching system implemented on a Field-Programmable Gate Array (FPGA), which is built around a lightweight, hardware-optimized dual-path Semi-Global Matching (SGM) algorithm. The proposed method simplifies the traditional eight-path cost aggregation into horizontal and vertical dual-path aggregation, substantially reducing hardware resource consumption while preserving matching accuracy. The system employs a pipelined architecture that integrates image capture, DDR3 caching, and HDMI display output. Experimental results demonstrate that under the configuration of a 5 × 5 matching window and a disparity range of 64, the system generates stable disparity maps at 60 frames per second, with total power consumption below 2.2 W and FPGA core logic utilization under 30%. Compared to the conventional eight-path SGM, the dual-path strategy incurs only a marginal 6% increase in average bad pixel rate on standard stereo datasets while reducing Block RAM (BRAM) usage by approximately 30%. This achieves an effective practical balance between accuracy, computational efficiency, and power consumption.
Building similarity graph...
Analyzing shared references across papers
Loading...
Yang Song
Hongyu Sun
Wenmin Song
Electronics
Chengdu University of Technology
China Railway Group (China)
Institute of Navigation
Building similarity graph...
Analyzing shared references across papers
Loading...
Song et al. (Wed,) studied this question.
www.synapsesocial.com/papers/69d896a46c1944d70ce082b3 — DOI: https://doi.org/10.3390/electronics15081549