With　the　background　of　visual　servomechanism　for　robotic　arm　and　with　the　aim　of　features　required　by　stereo　vision,　a　method　for　real-time　feature　extraction　on　the　base　of　block　of　color　is　presented.　The　localization　of　feature　includes　coarse　location　and　fine　location.　The　coarse　location　aim　is　to　track　the　object.　The　fine　location　is　on　the　base　of　coarse　location,　cuts　a　small　image　from　the　origin　image,　extracts　the　contour　of　the　object　from　the　small　image　and　classifies　the　contour　by　K-L　transform,　then　fits　the　edge　points　by　Lease　Square　Error　(LSE)　method.　The　sub-pixel　coordinates　of　the　feature　points　are　calculated　by　the　equations　of　lines,　it　provides　foundation　for　the　measurement　of　object＇s　position　and　posture.　The　experimental　results　show　its　effectiveness.　©2009　IEEE.