A Hough Transform Algorithm Implementation Using Dynamic Parallelism With CUDA

ABSTRACT The Compute Unified Device Architecture (CUDA) parallel programming model has become very popular in image processing methods for detecting shapes. Dynamic parallelism with CUDA allows nesting and recursion algorithms to be implemented by using multiple child kernels created dynamically from their parent kernels. This technique allows the computational resources associated with the Graphics Processing Units (GPUs) to be used more efficiently, providing homogeneous workloads between threads and blocks. This paper proposes the use of dynamic parallelism applied to the image decomposition of the Hough transform to detect straight lines. The comparison results with another parallel algorithm published in the literature, which does not use dynamic parallelism, are presented in terms of time, reducing the total execution time and obtaining a performance of up to 1.7 times faster.