Edge　detection　is　an　important　research　field　in　image　processing　and　computer　vision,　especially　in　feature　extraction.　The　detection　efficiency　of　the　traditional　Sobel　edge　detection　algorithm　is　reduced　due　to　serial　instructions,　and　its　recognition　ability　for　complex　image　textures　is　weak.　The　threshold　needs　to　be　set　artificially.　In　this　context,　we　present　an　improved　Sobel　algorithm,　which　enlarges　the　traditional　operator　convolution　factor　size　to　5×5,　extends　the　direction　to　16　directions,　and　introduces　the　median　as　a　local　adaptive　threshold　for　judgment.　The　improved　algorithm　is　implemented　using　FPGA　and　MATLAB　respectively,　and　processes　image　data　with　a　resolution　of　1280×720.　The　experimental　results　show　that:　When　the　clock　frequency　of　the　image　processing　module　is　84　MHz,　the　FPGA　processes　a　1280×720　image　in　about　0.0109　s,　while　the　MATLAB　processes　the　same　image　in　an　average　time　of　0.915　s.　The　Sobel　edge　detection　algorithm　proposed　in　this　work　effectively　denoises　the　image　while　also　detecting　the　edge　information　of　the　image　efficiently　and　accurately.　The　algorithm　processing　speed　is　improved　by　about　84　times,　which　can　well　meet　the　real-time　requirements　of　video　image　processing　systems.　©　2024　John　Wiley　and　Sons　Inc.　All　rights　reserved.