【植物检测】基于对称的作物田三维点云植物检测研究（Matlab代码实现）

​

欢迎来到本博客❤️❤️

博主优势： 博客内容尽量做到思维缜密，逻辑清晰，为了方便读者。r/> ⛳️座右铭：行百里者，半于九十。

精准农业中使用的机器人系统的关键要求之一是能够检测单个植物的确切位置。 我们提出了一种基于从种植园获得的3D点云的新方法，该方法确定确切的植物位置，可用于基于地标的SLAM（同时定位） 和映射）在移动机器人上或用于引导工厂处理系统。 我们方法背后的关键思想是利用同一物种的单个植物之间的几何相似性以及它们播种的规则模式。 因此，我们的方法对于作物以规则模式排列的单一栽培种植园最有用。 在我们的实验研究中，我们表明所提出的方法优于对不同生长阶段记录的真实世界数据进行的传统平均偏移分析。

部分代码：

%% plot examples from all daysfor day = 1:nDays h = figure(day+798); datasets = days(day).datasets; data = load(strcat(basepathData, datasets{2})); pNorm = geometry.normalizePointCloud(data.mapFull); plot.pc(pNorm(:,1:2), data.mapColorFull, 'markersize', 20); ax = gca; ax.FontSize = 5; ax.XLim = [-3 3]; ax.YLim = [-0.3 0.3]; set(gcf, 'Position', [200.6040 713.8713 691.0099 120.7129]) print(h, '-dpng', [basepathImages 'example_day' num2str(day) '.png'], '-r300')end

%% do the evaluation for each error_thresholdfor threshold_error = [0.01 0.02 0.04]

%% for day = 1:nDays %% evaluate each day individually datasets = days(day).datasets;

% combine individual datasets for each day dayCombined.landmarksMeanshift = []; dayCombined.landmarksMeanshiftScores = []; dayCombined.landmarksApproach = []; dayCombined.landmarkScores = []; dayCombined.landmarksManual = []; dayCombined.map = []; dayCombined.mapColor = [];

for j = 1:numel(datasets) datasetPath = [basepathData datasets{j}]; data = load(datasetPath); %% plant soil segmentation if ~isfield(data, 'map') [segmentation, extractorPolygon1, extractorPolygon2] = tools.greenextract(data.mapFull, data.mapColorFull); data.map = data.mapFull(segmentation,:); data.mapColor = data.mapColorFull(segmentation,:); save(datasetPath, '-struct', 'data', 'map', 'mapColor','-append') end %% manual landmarks if ~isfield(data, 'landmarksManual') h = figure(733); plot.pc(data.map, data.mapColor, 'markerSize', 20) datacursormode on dcm_obj = datacursormode(gcf); waitfor(msgbox('Click ok when finished labeling plants!')); cursor_info = getCursorInfo(dcm_obj); data.landmarksManual = vertcat(cursor_info.Position); if size(data.landmarksManual,2) == 3 data.landmarksManual = data.landmarksManual(:,1:2); end save(datasetPath, '-struct', 'data', 'landmarksManual','-append') end if updateLandmarksManual figure(733); h = plot.pc(data.map(:,1:2), data.mapColor, 'markerSize', 20); datacursormode on dcm_obj = datacursormode(gcf); hTarget = handle(h); xdata = get(hTarget,'XData'); ydata = get(hTarget,'YData'); idx = knnsearch([xdata' ydata'], data.landmarksManual);

for tipIdx = 1:size(data.landmarksManual, 1) hDatatip = dcm_obj.createDatatip(hTarget); propPointDataCursor = get(hDatatip,'Cursor'); % Move the datatip propPointDataCursor.DataIndex = idx(tipIdx); pos = [xdata(idx(tipIdx)) ydata(idx(tipIdx))]; propPointDataCursor.Position = pos; set(hDatatip,'Position',pos) end waitfor(msgbox('Click ok when finished labeling plants!')); cursor_info = getCursorInfo(dcm_obj); data.landmarksManual = vertcat(cursor_info.Position); if size(data.landmarksManual,2) == 3 data.landmarksManual = data.landmarksManual(:,1:2); end save(datasetPath, '-struct', 'data', 'landmarksManual','-append') end %% tf density approach on whole map if recalculateApproach && isfield(data, 'landmarksApproach') data = rmfield(data, 'landmarksApproach'); end if ~isfield(data, 'landmarksApproach') [data.landmarksApproach, data.landmarkScores] = optimization.detectPlants(data.map, days(day).args{:}); save(datasetPath, '-struct', 'data', 'landmarksApproach', 'landmarkScores', '-append') end %% mean shift baseline if recalculateMeanshift && isfield(data, 'landmarksMeanshift') data = rmfield(data, 'landmarksMeanshift'); end if ~isfield(data, 'landmarksMeanshift') [data.landmarksMeanshift,~,clusterSizes] = clustering.meanshift(data.map', days(day).args{12}); data.landmarksMeanshiftScores = vector.normalize(cellfun('length',clusterSizes)); save(datasetPath, '-struct', 'data', 'landmarksMeanshift', 'landmarksMeanshiftScores', '-append') end

%% combine dayCombined.landmarksMeanshift = [dayCombined.landmarksMeanshift; data.landmarksMeanshift]; dayCombined.landmarksMeanshiftScores = [dayCombined.landmarksMeanshiftScores; data.landmarksMeanshiftScores]; dayCombined.landmarksApproach = [dayCombined.landmarksApproach; data.landmarksApproach]; dayCombined.landmarkScores = [dayCombined.landmarkScores; data.landmarkScores]; dayCombined.landmarksManual = [dayCombined.landmarksManual; data.landmarksManual]; dayCombined.map = [dayCombined.map; data.map]; dayCombined.mapColor = [dayCombined.mapColor; data.mapColor]; end landmarksMeanshift = dayCombined.landmarksMeanshift; landmarksMeanshiftScores = dayCombined.landmarksMeanshiftScores; landmarksApproach = dayCombined.landmarksApproach; landmarkScores = dayCombined.landmarkScores; landmarksManual = dayCombined.landmarksManual; fprintf('day %d: %d labels\n', day,size(landmarksManual,1))

%% calculate errors [errorMeanshiftVecidx, errorMeanshift] = knnsearch(landmarksManual, landmarksMeanshift(:,1:2)); [errorApproachVecidx, errorApproach] = knnsearch(landmarksManual, landmarksApproach(:,1:2)); errorMeanshiftVec = landmarksManual(errorMeanshiftVecidx,:) - landmarksMeanshift(:,1:2); errorApproachVec = landmarksManual(errorApproachVecidx,:) - landmarksApproach(:,1:2);

%% eval approach figure(day); view(2) thresholds = []; rms = []; sizes = []; for step = 1:100 landmarksApproachSel = landmarksApproach(landmarkScores > ((step-1)/100),:); [errorApproachVecidx, errorApproach] = knnsearch(landmarksManual, landmarksApproachSel(:,1:2)); sizes(step) = size(landmarksApproachSel,1); thresholds(step) = step/100; rms(step) = sqrt(mean(errorApproach.^2)); if thresholds(step) == 0.9 fprintf('day %d, score threshold 0.9, rms %f, detection ratio %f\n', day, rms(step), sizes(step)/size(landmarksManual,1)) end end yyaxis left plot(thresholds,1000*rms, '--b') xlabel('score threshold')

[1]Janosch Dobler, Alexander Schaefer, Wolfram Burgard (2018) Symmetry-Based Plant Detection in 3D Point Clouds from Crop Fields

​