halcon機器視覺深度學習對象檢測，物體檢測

效果圖

在這里插入圖片描述

操作步驟

首先要在Deep Learning Tool工具里面把圖片打上標注文本，
然后訓練模型，導出模型文件

這個是模型
model_訓練-250215-111516_opt.hdl

模型配置參數
model_訓練-250215-111516_opt_dl_preprocess_params.hdict

軟件版本

使用的版本 halcon 23.11
Deep Learning Tool-24.05.1

halcon參考代碼

 
* 
* Inference can be done on a GPU or CPU.
* See the respective system requirements in the Installation Guide.
* If possible a GPU is used in this example.
* In case you explicitly wish to run this example on the CPU,
* choose the CPU device instead.
query_available_dl_devices (['runtime', 'runtime'], ['gpu', 'cpu'], DLDeviceHandles)
if (|DLDeviceHandles| == 0)throw ('No supported device found to continue this example.')
endif
* Due to the filter used in query_available_dl_devices, the first device is a GPU, if available.
*第一個設備是 GPU（如果可用）
DLDevice := DLDeviceHandles[0]
* * *************************************************
* **   設置推理路徑和參數   ***
* *************************************************
* 
* 我們將對示例圖像進行推理。
* 在實際應用程序中，新傳入的圖像（不用于訓練或評估）
* 將在此處使用。
* 
* 在此示例中，我們從 file 中讀取圖像。* 用我訓練的圖片
ImageDir :=  'G:/機器視覺_測試項目/家具目標檢測/images - 副本'* 
* Set the paths of the retrained model and the corresponding preprocessing parameters.
* Example data folder containing the outputs of the previous example series.
ExampleDataDir := 'detect_pills_data'* Use the pretrained model and preprocessing parameters shipping with HALCON.*使用 HALCON 附帶的預訓練模型和預處理參數。*PreprocessParamFileName := 'detect_pills_preprocess_param.hdict'* RetrainedModelFileName := 'detect_pills.hdl'*whl 測試我自己訓練的模型和參數，圖片配置dir1223:='G:/機器視覺_測試項目/家具目標檢測/'imgConfigHdict:='model_訓練-250215-111516_opt_dl_preprocess_params.hdict'PreprocessParamFileName:= dir1223+imgConfigHdict*識別模型* RetrainedModelFileName := dir1223+ 'best_model.hdl'RetrainedModelFileName :=dir1223+ 'model_訓練-250215-111516_opt.hdl'* 
* Batch Size used during inference.推理批次大小
BatchSizeInference := 1
* 
* Postprocessing parameters for the detection model.檢測模型的后處理參數。
MinConfidence := 0.6
MaxOverlap := 0.2
MaxOverlapClassAgnostic := 0.7
* 
* ********************
* **   推理   ***
* ********************
* 
* Check if all necessary files exist.
*check_data_availability (ExampleDataDir, PreprocessParamFileName, RetrainedModelFileName, UsePretrainedModel)
* 
*  讀取重新訓練的模型。
read_dl_model (RetrainedModelFileName, DLModelHandle)
* 
* Set the batch size. 設置批處理大小。
set_dl_model_param (DLModelHandle, 'batch_size', BatchSizeInference)
* 
* Initialize the model for inference.初始化模型以進行推理。
set_dl_model_param (DLModelHandle, 'device', DLDevice)
* 
* Set postprocessing parameters for model.設置模型的后處理參數。
set_dl_model_param (DLModelHandle, 'min_confidence', MinConfidence)
set_dl_model_param (DLModelHandle, 'max_overlap', MaxOverlap)
set_dl_model_param (DLModelHandle, 'max_overlap_class_agnostic', MaxOverlapClassAgnostic)
* 
* Get the parameters used for preprocessing.獲取用于預處理的參數。
read_dict (PreprocessParamFileName, [], [], DLPreprocessParam)
* * 使用顯示所需的數據集參數創建字典。
DLDataInfo := dict{}
get_dl_model_param (DLModelHandle, 'class_names', ClassNames)* 目標對象，標簽名稱
DLDataInfo.class_names := ClassNames
get_dl_model_param (DLModelHandle, 'class_ids', ClassIDs)
DLDataInfo.class_ids := ClassIDs
* 設置可視化的通用參數。
GenParam := dict{scale_windows: 1.2,display_labels:true}*讀取目錄里面的若干圖片文件list_files (ImageDir, ['files' ], ImageFiles)*獲取圖片尺寸，whl測試read_image(img1,ImageFiles[0])        get_image_size (img1, Width, Height)* dev_open_window (1, 1, Width, Height, 'black', WindowID1)dev_open_window (1, 1, 900, 900*Height/(Width*1.0), 'black', WindowID1)*視頻文件讀取*grab_image_from_video()* open_framegrabber()* 讀取視頻幀
*grab_image_start([])
*grab_image(Image)
*grab_image_stop([])* 
* 以 BatchSizeInference 大小批量循環訪問所有圖像以進行推理
for BatchIndex := 0 to floor(|ImageFiles| / real(BatchSizeInference)) - 1 by 1* * Get the paths to the images of the batch.Batch := ImageFiles[BatchIndex * BatchSizeInference:(BatchIndex + 1) * BatchSizeInference - 1]* 讀取圖片read_image (ImageBatch, Batch)* * Generate the DLSampleBatch.gen_dl_samples_from_images (ImageBatch, DLSampleBatch)* * Preprocess the DLSampleBatch.preprocess_dl_samples (DLSampleBatch, DLPreprocessParam)* * 在 DLSampleBatch 上應用 DL 模型。apply_dl_model (DLModelHandle, DLSampleBatch, [], DLResultBatch)* * Postprocessing and visualization.后處理和可視化* Loop over each sample in the batch.循環處理批次中的每個樣品for SampleIndex := 0 to BatchSizeInference - 1 by 1* * Get sample and according results.獲取樣本和相應的結果。DLSample := DLSampleBatch[SampleIndex]DLResult := DLResultBatch[SampleIndex]* *whl測試
KeysForDisplay:='bbox_result'* * 顯示檢測結果.* dev_display_dl_data (DLSample, DLResult, DLDataInfo, 'bbox_result', GenParam, WindowHandleDict)*whl測試，ocr_detection_score_map_character* 
* This procedure displays the content of the provided DLSample and/or DLResult
* depending on the input string KeysForDisplay.
* DLDatasetInfo is a dictionary containing the information about the dataset.
* The visualization can be adapted with GenParam.
* 
* ** Set the default values: ***
Params := dict{}
* 
* Define the screen width when a new window row is started.
Params.threshold_width := 1024
* Since potentially a lot of windows are opened,
* scale the windows consistently.
Params.scale_windows := 0.8
* Set a font and a font size.
Params.font := 'mono'
Params.font_size := 14
* 
Params.line_width := 2
Params.map_transparency := 'cc'
Params.map_color_bar_width := 140
* 
* Define parameter values specifically for 3d_gripping_point_detection
Params.gripping_point_color := '#00FF0099'
Params.gripping_point_size := 6
Params.region_color := '#FF000040'
Params.gripping_point_map_color := '#83000080'
Params.gripping_point_background_color := '#00007F80'
* 
* Define parameter values specifically for anomaly detection
* and Global Context Anomaly Detection.
Params.anomaly_region_threshold := -1
Params.anomaly_classification_threshold := -1
Params.anomaly_region_label_color := '#40e0d0'
Params.anomaly_color_transparency := '40'
Params.anomaly_region_result_color := '#ff0000c0'
* 
* Define segmentation-specific parameter values.
Params.segmentation_max_weight := 0
Params.segmentation_draw := 'fill'
Params.segmentation_transparency := 'aa'
Params.segmentation_exclude_class_ids := []
* 
* Define bounding box-specific parameter values.
Params.bbox_label_color := '#000000' + '99'
Params.bbox_display_confidence := 1
Params.bbox_text_color := '#eeeeee'
* 
* By default, display a description on the bottom.
Params.display_bottom_desc := true
* 
* By default, show a legend with class IDs.
Params.display_legend := true
* 
* By default, show the anomaly ground truth regions.
Params.display_ground_truth_anomaly_regions := true
* 
* By default, show class IDs and color frames for classification ground truth/results.
Params.display_classification_ids := true
Params.display_classification_color_frame := true
* 
* By default, show class labels for detection ground truth/results.
Params.display_labels := true
* 
* By default, show direction of the ground truth/results instances for detection with instance_type 'rectangle2'.
Params.display_direction := true
* 
* By default, use color scheme 'Jet' for the heatmap display.
Params.heatmap_color_scheme := 'jet'
* ** Set user-defined values: ***
* 
* Overwrite default values by given generic parameters.
if (GenParam != [])get_dict_param (GenParam, 'keys', [], GenParamNames)for ParamIndex := 0 to |GenParamNames| - 1 by 1GenParamName := GenParamNames[ParamIndex]get_dict_param (Params, 'key_exists', GenParamName, KeyExists)if (not KeyExists)throw ('Unknown generic parameter: ' + GenParamName + '.')endifParams.[GenParamName] := GenParam.[GenParamName]endfor
endif
* 
if (|DLSample| > 1 or |DLResult| > 1)throw ('Only a single dictionary for DLSample and DLResult is allowed')
endif
* 
* Get the dictionary keys.
get_dict_param (DLSample, 'keys', [], SampleKeys)
if (DLResult != [])get_dict_param (DLResult, 'keys', [], ResultKeys)
endif
* 
* Get image ID if it is available.
get_dict_param (DLSample, 'key_exists', 'image_id', ImageIDExists)
if (ImageIDExists)get_dict_tuple (DLSample, 'image_id', ImageID)ImageIDString := 'image ID ' + ImageIDImageIDStringBraces := '(image ID ' + ImageID + ')'ImageIDStringCapital := 'Image ID ' + ImageID
elseImageIDString := ''ImageIDStringBraces := ImageIDStringImageIDStringCapital := ImageIDString
endif
* AdditionalGreenClassNames := []
KeyIndex := 0* whl添加if* 
* Check if DLDatasetInfo is valid.* whl添加
DLDatasetInfo:=DLDataInfo* Check if DLDatasetInfo contains necessary keys.ClassKeys := ['class_names', 'class_ids']get_handle_param (DLDatasetInfo, 'key_exists', ClassKeys, ClassKeysExist)if (min(ClassKeysExist) == 0)* In that case we expect that the class names and ids are never used.elseget_handle_param (DLDatasetInfo, 'keys', [], DLDatasetInfoKeys)for Index := 0 to |ClassKeys| - 1 by 1if (find_first(DLDatasetInfoKeys,ClassKeys[Index]) == -1)throw ('Key ' + ClassKeys[Index] + ' is missing in DLDatasetInfo.')endifendfor* * Get the general dataset information, if available.get_handle_tuple (DLDatasetInfo, 'class_names', ClassNames)get_handle_tuple (DLDatasetInfo, 'class_ids', ClassIDs)* * 為類定義不同的顏色*   get_dl_class_colors (ClassNames, AdditionalGreenClassNames, Colors)* 函數get_dl_class_colors 替代者，開始* Define distinct colors for the classes.
NumColors := |ClassNames|
* Get distinct colors without randomness makes neighboring colors look very similar.
* We use a workaround to get deterministic colors where subsequent colors are distinguishable.
get_distinct_colors (NumColors, false, 0, 200, ColorsRainbow)tuple_inverse (ColorsRainbow, ColorsRainbow)
make_neighboring_colors_distinguishable (ColorsRainbow, Colors)
* If a class 'OK','ok', 'good' or 'GOOD' or a class specified in AdditionalGreenClassNames is present set this class to green.
* Only the first occurrence found is set to a green shade.
tuple_union (['good', 'GOOD', 'ok', 'OK'], AdditionalGreenClassNames, ClassNamesGood)
for IndexFind := 0 to |ClassNamesGood| - 1 by 1GoodIdx := find_first(ClassNames,ClassNamesGood[IndexFind])if (GoodIdx != -1 and |ClassNames| <= 8)* If number of classes is <= 8, swap color with a green color.CurrentColor := Colors[GoodIdx]GreenIdx := floor(|ClassNames| / 2.0)* Set to pure green.Colors[GoodIdx] := '#00ff00'* Write original color to a green entry.Colors[GreenIdx] := CurrentColorbreakelseif (GoodIdx != -1 and |ClassNames| > 8)* If number of classes is larger than 8, set the respective color to green.Colors[GoodIdx] := '#00ff00'breakendif
endfor
* 函數get_dl_class_colors 替代者，結束endif* 
* ** Set window parameters: ***
* * 
* ** Display the data: ***
* 
* Display data dictionaries.
KeyIndex := 0*while (KeyIndex < |KeysForDisplay|)* * if (KeysForDisplay[KeyIndex] == 'bbox_result' or KeysForDisplay[KeyIndex] == 'ocr_detection_result')* * Result bounding boxes on image.圖像上的結果邊界框。get_dl_sample_image (Image, SampleKeys, DLSample, 'image')* get_dl_sample_image (ImageBatch, SampleKeys, DLSample, 'image')* * Get or open next window.訓練時的圖片寬高get_image_size (Image, WidthImage, HeightImage)* get_next_window (Params.font, Params.font_size, Params.display_bottom_desc, WidthImage, HeightImage, 0, Params.scale_windows, Params.threshold_width, PrevWindowCoordinates, WindowHandleDict, KeysForDisplay[KeyIndex], CurrentWindowHandle, WindowImageRatio, PrevWindowCoordinates)*原始代碼，whl測試注釋，訓練時的壓縮后圖片* dev_display (Image)*whl添加，獲取窗口尺寸* get_window_extents(WindowID1,Row,Column,Window_Width,Window_Height)*圖片原圖本身尺寸，非訓練設置壓縮的圖片尺寸get_image_size (ImageBatch, WidthBig, HeightBig)*whl添加，比值*應該先把訓練時圖片的原始框點轉換圖片本身尺寸時的坐標就可以了imgRate:=1imgHeightBeiWidth:=1if(1)*寬度，乘以1.0轉為小數，可以讓除得到小數結果imgRate:=WidthBig/(WidthImage*1.0)* 高度占寬度的比值imgHeightBeiWidth:=HeightBig/(HeightImage*1.0)endif*whl 添加，顯示原圖片* 調整圖像尺寸* zoom_image_size(ImageBatch,imgZoom,800,800*HeightBig/(WidthBig*1.0),'constant')* zoom_image_size(ImageBatch,imgZoom,800,800*HeightBig/(WidthBig*1.0),'constant')dev_clear_window()*whl 添加，顯示原圖片 dev_display(ImageBatch)*讓窗口適應圖片的尺寸，窗口跟圖片一樣大*   dev_resize_window_fit_image (ImageBatch, 0, 0, -1, -1)* dev_re* dev_open_window_fit_image (ImageBatch, 0, 0, -1, -1, WindowID1)*dev_resize_window_fit_size (0, 0, -1, -1, -1, -1)*full_domain(ImageBatch,ImageBatch）* dev_set_window(WindowID1)* dev_set_part*whl 添加測試WindowImageRatio:=1CurrentWindowHandle:=WindowID1*目標對象分類文本* className:=DLResult.bbox_class_name*顯示目標對象框 *dev_display_result_detection (DLResult, ResultKeys, Params.line_width, ClassIDs, TextConf, Colors, Params.bbox_label_color, WindowImageRatio, 'top', Params.bbox_text_color, Params.display_labels, DisplayDirectionTemp, CurrentWindowHandle, BboxClassIndex)*dev_display_result_detection (DLResult, ResultKeys, Params.line_width, ClassIDs, TextConf, Colors, Params.bbox_label_color, WindowImageRatio, 'top', Params.bbox_text_color, Params.display_labels, DisplayDirectionTemp, CurrentWindowHandle, BboxClassIndex)*目標文本顯示set_display_font (WindowID1, 12, 'mono', 'false', 'false')  *提取函數，顯示目標對象框，識別分類文本，開始
InstanceType := ''
MaskExists := false
if (find(ResultKeys,'bbox_row1') != -1)   *進這個get_dict_tuple (DLResult, 'bbox_row1', BboxRow1)get_dict_tuple (DLResult, 'bbox_col1', BboxCol1)get_dict_tuple (DLResult, 'bbox_row2', BboxRow2)get_dict_tuple (DLResult, 'bbox_col2', BboxCol2)InstanceType := 'rectangle1'*1進入，0不進入if(1)     *whl 添加，乘以系數*高度BboxRow1:=BboxRow1*imgHeightBeiWidth        BboxRow2:=BboxRow2*imgHeightBeiWidth*寬度BboxCol1:=BboxCol1*imgRateBboxCol2:=BboxCol2*imgRate*whl 添加，重置為1imgRate:=1endifelseif (find(ResultKeys,'bbox_phi') != -1)get_dict_tuple (DLResult, 'bbox_row', BboxRow)get_dict_tuple (DLResult, 'bbox_col', BboxCol)get_dict_tuple (DLResult, 'bbox_length1', BboxLength1)get_dict_tuple (DLResult, 'bbox_length2', BboxLength2)get_dict_tuple (DLResult, 'bbox_phi', BboxPhi)get_dict_tuple (DLResult, 'bbox_class_id', BboxClasses)InstanceType := 'rectangle2'
elsethrow ('Result bounding box data could not be found in DLResult.')
endif
if (find(ResultKeys,'mask') != -1)get_dict_object (InstanceMask, DLResult, 'mask')MaskExists := true
endif
if (InstanceType != 'rectangle1' and InstanceType != 'rectangle2' and not MaskExists)throw ('Result bounding box or mask data could not be found in DLSample.')
endif*whl注釋
get_dict_tuple (DLResult, 'bbox_class_id', BboxClasses)* whl 添加，顯示檢測對象名稱*whl添加
ShowLabels:=true
ShowDirection:=true
TextColor:='#eeeeee'TextConf:=''if (|BboxClasses| > 0)* * Get text and text size for correct positioning of result class IDs.if (ShowLabels)Text := BboxClasses + TextConfget_string_extents (CurrentWindowHandle, Text, Ascent, Descent, _, _)TextOffset := (Ascent + Descent) / WindowImageRatioendif* * Generate bounding box XLDs.if (InstanceType == 'rectangle1')tuple_gen_const (|BboxRow1|, 0.0, BboxPhi)*畫目標框線，乘以 imgRategen_rectangle2_contour_xld (BboxRectangle, 0.5 * (BboxRow1 + BboxRow2), 0.5 * (BboxCol1 + BboxCol2), BboxPhi, 0.5 * (BboxCol2 - BboxCol1), 0.5 * (BboxRow2 - BboxRow1))* gen_rectangle2_contour_xld (BboxRectangle, 0.5 * (BboxRow1  + BboxRow2)*imgRate, 0.5 * (BboxCol1 + BboxCol2)*imgRate, BboxPhi, 0.5 * (BboxCol2 - BboxCol1)*imgRate, 0.5 * (BboxRow2 - BboxRow1)*imgRate)if (ShowLabels)LabelRowTop := BboxRow1LabelRowBottom := BboxRow2 - TextOffsetLabelCol := BboxCol1endifelseif (InstanceType == 'rectangle2')gen_rectangle2_contour_xld (BboxRectangle, BboxRow, BboxCol, BboxPhi, BboxLength1, BboxLength2)if (ShowLabels)LabelRowTop := BboxRow - TextOffsetLabelRowBottom := BboxRowLabelCol := BboxColendifif (ShowDirection)if (ShowDirection == -1)ArrowSizeFactorLength := 0.4ArrowSizeFactorHead := 0.2MaxLengthArrow := 20HalfLengthArrow := min2(MaxLengthArrow,BboxLength1 * ArrowSizeFactorLength)ArrowBaseRow := BboxRow - (BboxLength1 - HalfLengthArrow) * sin(BboxPhi)ArrowBaseCol := BboxCol + (BboxLength1 - HalfLengthArrow) * cos(BboxPhi)ArrowHeadRow := BboxRow - (BboxLength1 + HalfLengthArrow) * sin(BboxPhi)ArrowHeadCol := BboxCol + (BboxLength1 + HalfLengthArrow) * cos(BboxPhi)ArrowHeadSize := min2(MaxLengthArrow,min2(BboxLength1,BboxLength2)) * ArrowSizeFactorHeadelseArrowHeadSize := 20.0ArrowBaseRow := BboxRowArrowBaseCol := BboxColArrowHeadRow := BboxRow - (BboxLength1 + ArrowHeadSize) * sin(BboxPhi)ArrowHeadCol := BboxCol + (BboxLength1 + ArrowHeadSize) * cos(BboxPhi)endifgen_arrow_contour_xld (OrientationArrows, ArrowBaseRow, ArrowBaseCol, ArrowHeadRow, ArrowHeadCol, ArrowHeadSize, ArrowHeadSize)endifelseif (MaskExists)area_center (InstanceMask, _, MaskRow, MaskCol)LabelRowTop := MaskRow - TextOffsetLabelRowBottom := MaskRowLabelCol := MaskColelsethrow ('Unknown instance_type: ' + InstanceType)endif* get_contour_style (CurrentWindowHandle, ContourStyle)dev_set_contour_style ('stroke')get_line_style (CurrentWindowHandle, Style)*whl添加LineWidthBbox:=1LineWidths := [LineWidthBbox + 2,LineWidthBbox]dev_set_line_width (LineWidthBbox)* * Collect ClassIDs of the bounding boxes.tuple_gen_const (|BboxClasses|, 0, BboxClassIndices)* * Draw bounding boxes.for IndexBbox := 0 to |BboxClasses| - 1 by 1ClassID := find(ClassIDs,BboxClasses[IndexBbox])BboxClassIndices[IndexBbox] := ClassID* First draw in black to make the class-color visible.CurrentColors := ['black',Colors[ClassID]]if (MaskExists)select_obj (InstanceMask, MaskSelected, IndexBbox + 1)dev_set_draw ('fill')dev_set_color (Colors[ClassID] + '80')dev_display (MaskSelected)dev_set_draw ('margin')endiffor IndexStyle := 0 to |CurrentColors| - 1 by 1dev_set_color (CurrentColors[IndexStyle])dev_set_line_width (LineWidths[IndexStyle])if (InstanceType != '')select_obj (BboxRectangle, RectangleSelected, IndexBbox + 1)dev_display (RectangleSelected)if (InstanceType == 'rectangle2' and ShowDirection)select_obj (OrientationArrows, ArrowSelected, IndexBbox + 1)dev_display (ArrowSelected)endifendifendforendfor* * Draw text of bounding boxes.if (ShowLabels)* For better visibility the text is displayed after all bounding boxes are drawn.* Get text and text size for correct positioning of result class IDs.* Text := BboxClasses + TextConf*whl 對象文本*bbox_class_name標簽，bbox_confidence置信度得分whlObjectClassName:=DLResult.bbox_class_name*四舍五入，保留10位小數tuple_string(DLResult.bbox_confidence, '.10f', StringConfidence)     *截取字符串tuple_substr (StringConfidence, 0, 3, Substring)Text :=whlObjectClassName+ Substring* Select text color.if (TextColor == '')TextColorClasses := Colors[BboxClassIndices]elsetuple_gen_const (|BboxClassIndices|, TextColor, TextColorClasses)endif* Select correct position of the text.LabelRow := LabelRowTop*whl注釋
*         if (TextPositionRow == 'bottom')*     LabelRow := LabelRowBottom* endif*whl添加，標簽字體背景色BoxLabelColor:='#00000099'  * BoxLabelColor:='#05E600'* Display text.顯示對象標簽文本         dev_disp_text (Text, 'image', LabelRow, LabelCol, TextColorClasses, ['box_color', 'shadow', 'border_radius'], [BoxLabelColor,'false', 0])endif* dev_set_contour_style (ContourStyle)set_line_style (CurrentWindowHandle, Style)
else* Do nothing if no results are present.BboxClassIndices := []
endif*顯示目標對象框，識別分類文本，結束*whl 注釋，不執行if代碼里面的代碼endif* KeyIndex := KeyIndex + 1
*endwhile      * whl測試，目標框顯示，結束             *whl注釋* WindowHandles := WindowHandleDict.bbox_result* dev_set_window (WindowHandles[0])* set_display_font (WindowHandles[0], 16, 'mono', 'true', 'false')* whl測試* set_display_font (WindowID1, 16, 'mono', 'true', 'false')*whl注釋，不顯示綠色的檢測文本列表*  dev_disp_text (Text, 'window', 'top', 'left', TextColor, ['box_color', 'shadow'], [TextBoxColor,'false'])set_display_font (WindowID1, 16, 'mono', 'true', 'false')  * dev_disp_text ('Press Run (F5) to continue', 'window', 'bottom', 'right', 'black', [], [])* 拆分字符串，圖片路徑     tuple_split(Batch,'\\',fileWordArr)Wordlength:=|fileWordArr|*取最后一個字符串fileShortName:=fileWordArr[Wordlength-1]*顯示文件名dev_disp_text (fileShortName, 'window', 'bottom', 'left', 'magenta', [], [])*將窗口保存為本地圖片文件* dump_window(WindowID1,'png','G:/機器視覺_測試項目/家具目標檢測/videoImages/2')stop ()endfor
endfor
* 
* Close windows used for visualization.關閉用于可視化的窗口
*dev_close_window_dict (WindowHandleDict)
* 
* 
set_display_font (WindowID1, 24, 'mono', 'true', 'false')       dev_disp_text ('程序結束', 'window', 'bottom', 'right', 'green', ['box_color'], [ 'blue'])

本地函數 get_distinct_colors()

* 
* We get distinct color-values first in HLS color-space.
* Assumes hue [0, EndColor), lightness [0, 1), saturation [0, 1).
* 
* Parameter checks.
* NumColors.
if (NumColors < 1)throw ('NumColors should be at least 1')
endif
if (not is_int(NumColors))throw ('NumColors should be of type int')
endif
if (|NumColors| != 1)throw ('NumColors should have length 1')
endif
* Random.
if (Random != 0 and Random != 1)tuple_is_string (Random, IsString)if (IsString)Random := Random == 'true' or 'false'elsethrow ('Random should be either true or false')endif
endif
* StartColor.
if (|StartColor| != 1)throw ('StartColor should have length 1')
endif
if (StartColor < 0 or StartColor > 255)throw ('StartColor should be in the range [0, 255]')
endif
if (not is_int(StartColor))throw ('StartColor should be of type int')
endif
* EndColor.
if (|EndColor| != 1)throw ('EndColor should have length 1')
endif
if (EndColor < 0 or EndColor > 255)throw ('EndColor should be in the range [0, 255]')
endif
if (not is_int(EndColor))throw ('EndColor should be of type int')
endif
* 
* Color generation.
if (StartColor > EndColor)EndColor := EndColor + 255
endif
if (NumColors != 1)Hue := (StartColor + int((EndColor - StartColor) * real([0:NumColors - 1]) / real(NumColors - 1))) % 255
elseHue := mean([StartColor,EndColor])
endif
if (Random)Hue := Hue[sort_index(rand(NumColors))]Lightness := int((5.0 + rand(NumColors)) * 255.0 / 10.0)Saturation := int((9.0 + rand(NumColors)) * 255.0 / 10.0)
elseLightness := int(gen_tuple_const(NumColors,0.55) * 255.0)Saturation := int(gen_tuple_const(NumColors,0.95) * 255.0)
endif
* 
* Write colors to a 3-channel image in order to transform easier.
gen_image_const (HLSImageH, 'byte', 1, NumColors)
gen_image_const (HLSImageL, 'byte', 1, NumColors)
gen_image_const (HLSImageS, 'byte', 1, NumColors)
get_region_points (HLSImageH, Rows, Columns)
set_grayval (HLSImageH, Rows, Columns, Hue)
set_grayval (HLSImageL, Rows, Columns, Lightness)
set_grayval (HLSImageS, Rows, Columns, Saturation)
* 
* Convert from HLS to RGB.
trans_to_rgb (HLSImageH, HLSImageL, HLSImageS, ImageR, ImageG, ImageB, 'hls')
* 
* Get RGB-values and transform to Hex.
get_grayval (ImageR, Rows, Columns, Red)
get_grayval (ImageG, Rows, Columns, Green)
get_grayval (ImageB, Rows, Columns, Blue)
Colors := '#' + Red$'02x' + Green$'02x' + Blue$'02x'
return ()
*

本地函數 make_neighboring_colors_distinguishable()

* 
* Shuffle the input colors in a deterministic way
* to make adjacent colors more distinguishable.
* Neighboring colors from the input are distributed to every NumChunks
* position in the output.
* Depending on the number of colors, increase NumChunks.
NumColors := |ColorsRainbow|
if (NumColors >= 8)NumChunks := 3if (NumColors >= 40)NumChunks := 6elseif (NumColors >= 20)NumChunks := 4endifColors := gen_tuple_const(NumColors,-1)* Check if the Number of Colors is dividable by NumChunks.NumLeftOver := NumColors % NumChunksColorsPerChunk := int(NumColors / NumChunks)StartIdx := 0for S := 0 to NumChunks - 1 by 1EndIdx := StartIdx + ColorsPerChunk - 1if (S < NumLeftOver)EndIdx := EndIdx + 1endifIdxsLeft := [S:NumChunks:NumColors - 1]IdxsRight := [StartIdx:EndIdx]Colors[S:NumChunks:NumColors - 1] := ColorsRainbow[StartIdx:EndIdx]StartIdx := EndIdx + 1endfor
elseColors := ColorsRainbow
endif
return ()