FAST特征点检测算法的matalb源码实现

%this class is to realize FAST feature extractor classdef myFAST properties numInChain = 12; threshold = 0.3; img; featureMap; end methods function obj = myFAST(img_) if ndims(img_) == 3 obj.img = rgb2gray(img_); else obj.img = img_; end end % FAST的实现 function [corners] = operate(obj) corners = []; % 特征点 tempCorners = []; % 特征点的candidate [m, n] = size(obj.img); obj.featureMap = zeros(m, n); %特征点的打分值，如果不是特征点打分为0 for i = 4:m-4 for j = 4:n-4 % p0为检测点 p0 = obj.img(i,j); % 依次提取当前点半径为3的Bresenham圆圆周上的点，检测点正上方为p（1） p = []; p(end + 1) = obj.img(i - 3 ,j ); %p1 p(end + 1) = obj.img(i - 3 ,j + 1); %p2 p(end + 1) = obj.img(i - 2 ,j + 2); %p3 p(end + 1) = obj.img(i - 1 ,j + 3); %p4 p(end + 1) = obj.img(i ,j + 3); %p5 p(end + 1) = obj.img(i + 1 ,j + 3); %p6 p(end + 1) = obj.img(i + 2 ,j + 2); %p7 p(end + 1) = obj.img(i + 3 ,j + 1); %p8 p(end + 1) = obj.img(i + 3 ,j ); %p9 p(end + 1) = obj.img(i + 3 ,j - 1); %p10 p(end + 1) = obj.img(i + 2 ,j - 2); %p11 p(end + 1) = obj.img(i + 1 ,j - 3); %p12 p(end + 1) = obj.img(i ,j - 3); %p13 p(end + 1) = obj.img(i - 1 ,j - 3); %p14 p(end + 1) = obj.img(i - 2 ,j - 2); %p15 p(end + 1) = obj.img(i - 3 ,j - 1); %p16 thr = obj.threshold*p0; %判断该点是否为特征点的candidate %featureFlag = 1 代表周围有n个以上的点大于p0 + thr %featureFlag = 2 代表周围有n个以上的点小于p0 - thr featureFlag = FastFeaturePointDetect(p0,p,thr,obj.numInChain); if featureFlag == 1 || featureFlag == 2 %如果是特征点，计算特征点的打分值 val = ExtremeSuppression(p0,p,thr, obj.numInChain,featureFlag); tempCorners(end + 1).x = j; tempCorners(end).y = i; tempCorners(end).val = val; obj.featureMap(i,j) = val; end end end %极值点抑制 for k = 1:length(tempCorners) j = tempCorners(k).x; i = tempCorners(k).y; tempList(1) = obj.featureMap(i - 1,j - 1); tempList(2) = obj.featureMap(i - 1,j ); tempList(3) = obj.featureMap(i - 1,j + 1); tempList(4) = obj.featureMap(i ,j - 1); tempList(5) = obj.featureMap(i ,j + 1); tempList(6) = obj.featureMap(i - 1,j - 1); tempList(7) = obj.featureMap(i - 1,j ); tempList(8) = obj.featureMap(i - 1,j + 1); %如果特征点的打分值是附近3*3区域里最大的则将tempCorners（k）付给corners if tempCorners(k).val > max(tempList) corners(end+1).x = tempCorners(k).x; corners(end).y = tempCorners(k).y; corners(end).val = tempCorners(k).val; end end end end end %判断该点是否为特征点的candidate function flag = FastFeaturePointDetect(p0,p,thr, numInChain) flag = 0; % 遍历16个点，看是否有numInChain个点大于p0 + thr % 循环列表的index = mod(k-1,length(p))+1 for i = 1:16 if p(i) - p0 > thr counter = 1; for j = 1:numInChain - 1 k = i + j; if p(mod(k-1,length(p))+1) - p0 > thr counter = counter + 1; else break; end end if counter == numInChain flag = 1; return end else continue; end end % 遍历16个点，看是否有numInChain个点小于p0 - thr for i = 1:16 if p0 - p(i) > thr counter = 1; for j = 1:numInChain - 1 k = i + j; if p0 - p(mod(k-1,length(p))+1) > thr counter = counter + 1; else break; end end if counter == numInChain flag = 2; return end else continue; end end end %给特征点打分 function [val] = ExtremeSuppression(p0, p,thr, numInChain,featureFlag) N = length(p); if featureFlag == 1 d = p - double(p0); else d = double(p0) - p; end a0 = thr; %打分的结果一定是d（1）到d(16)中的一个，遍历16次找出最大的满足条件的d(k) for i = 1:16 a = min([d(mod(i,N)+1),d(mod(i+1,N)+1),d(mod(i+2,N)+1)]); if a < a0 continue; end for j = 4:numInChain-1 a = min(a,d(mod(i+j-1,N)+1)); end a0 = max(a0,min(a, d(mod(i-1,N)+1) )); a0 = max(a0, min(a, d(mod(i-1,N)+1) )); end val = a0 -1; end