Extension and FPGA architecture of the Generalized Hough Transform for real-time stereo correspondence

The Generalized Hough Transform (GHT) is a widely known image processing method to find arbitrary shapes in images. In the first part of this work, we present an extension of the Hough voting scheme to perform the disparity calculation of stereo correspondence. The approach is optimized for a hardware implementation in terms of processing regularity, data storage requirements and pipelining. The second part of this work is the presentation of a raster scan architecture of the main processing blocks in an FPGA. The architecture employs high data throughput with low latency and low resource utilization. Compared to other approaches, our method achieves comparable results in terms of disparity information calculation performance, tested with real world benchmark datasets. The hardware architecture outperforms other approaches in terms of speed and resource utilization.