Salient Line Extraction Using Symmetric Convolution for Embedded Vision

The patent segments the information sign detection task into distinct processing stages: gradient computation, non-maximum suppression, and thresholding. Each stage processes specific features independently, allowing parallel computation and avoiding the exhaustive pixel-by-pixel analysis of sliding window methods, thus improving computation speed while maintaining detection accuracy.

The patent performs preliminary gradient computation and directional analysis before final detection. By pre-calculating gradient magnitudes and directions, and identifying candidate regions through non-maximum suppression in advance, the system reduces the computational burden during real-time detection, achieving both high speed and accuracy.

The patent merges multiple processing operations into unified computational steps. The gradient computation stage simultaneously calculates gradient magnitude and direction, the non-maximum suppression stage combines thresholding and peak detection, reducing the number of separate processing modules and simplifying the overall system architecture while maintaining multi-stage processing benefits.

The patent designs processing modules that perform multiple functions. The gradient computation module serves both edge detection and orientation analysis, the non-maximum suppression module performs both thresholding and local maxima identification. This multi-functionality reduces the total number of components needed, decreasing device complexity while preserving recognition accuracy.

The patent replaces the mechanical sliding window approach with a field-based gradient computation method. Instead of physically moving a window across the image, the system computes gradients at all pixels simultaneously using convolution with gradient kernels, enabling parallel processing on embedded systems and dramatically improving detection speed while maintaining compatibility with standard image processing architectures.

The patent implements a periodic processing pipeline where gradient computation, non-maximum suppression, and thresholding are performed in sequential passes through the image data. This periodic structure allows embedded systems to process images in manageable stages, improving throughput and productivity while maintaining system compatibility through standardized processing intervals.