PIECEWISE POLYNOMIAL APPROXIMATION METHOD FOR CONVOLUTION WITH LARGE KERNEL

Image processing became a vital role in autonomous driving systems. In recent years, high-resolution cameras are starting to gain popularity and economically feasible to mount on mass-produced cars. To utilize all data from high-resolution cameras, we need to have large image kernels, which need to be processed on real-time embedded systems. For this purpose, we introduce new numerical methods for convolutional computation using piecewise polynomials approximation on filters. The computation of these methods does not increase by the size of the kernels, because the complexity depends on the control points of approximation. Also, it does not require additional hardware and does not have a limitation on memory size. Hence, our method is a good alternative for computing convolution with large filters in real-time embedded systems.