A new hardware architecture of high-performance real-time texture classification system based on FPGA

The visual system is essential as a critical source for intelligent robots to acquire external information. Nevertheless, the real-time performance of existing approaches remains inadequate. To address this, a new high-performance target classification system based on FPGA has been developed as part of the visual system. This system optimizes the hardware architecture of the target classification algorithm, incorporating a novel method aimed at boosting parallelism to improve real-time performance. The system is implemented on the Xilinx Zynq-7045 FPGA. Experimental results demonstrate that, for a grayscale image with a resolution of 128 \(\times\) $\times$ 128, the feature extraction time is merely 85.64 µs, achieving a speed three orders of magnitude greater than that of the MATLAB platform. Additionally, the resource consumption of this design is lower than that of existing hardware architectures.