halcon例程解析：数回收箱箱里的瓶子——check_bottle_crate

一路前行，幸运相伴

已于 2022-04-22 10:42:58 修改

阅读量1k

点赞数 5

分类专栏： halcon 文章标签：计算机视觉图像处理图像识别

于 2021-07-03 16:53:43 首次发布

本文链接：https://blog.youkuaiyun.com/ShareProgress/article/details/118442294

版权

halcon 专栏收录该内容

10 篇文章

订阅专栏

该博客介绍了如何使用Halcon算法来计数回收箱内的瓶子。通过漫射光照明，结合简单的阈值处理和形态学操作，能够有效地分割出瓶子和异物。首先，通过亮度阈值分割背景和瓶子区域，再去除大异物和噪声。接着，利用形状特征如宽、高、圆度进一步精确定位瓶子。最后，识别倒置的瓶子并进行结果显示。此方法对于倒置和横放的瓶子都有较好的识别能力。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

This example counts bottles in bottle crates of returned empties. With diffuse front light, correctly inserted bottles can be segmented relatively easy with a simple thresholding operation, followed by some basic morphology operations. Bottles inserted upside-down are also detected and displayed in orange, while across lying bottles or big objects are marked as clutter and produce a warning.

这个例子数的是回收箱中的空瓶子。使用漫射前光，正确插入的瓶子可以相对容易地分割，只需简单的阈值操作，然后进行一些基本的形态学操作。倒插的瓶子也能被检测到，并显示为橙色，而横放的瓶子或大物体则被标记为杂物，并产生警告。

1. 效果展示

在这里插入图片描述

2. 思路分析

这个问题比较简单，不需要太复杂的算法，只通过亮度等进行阈值分割、面积圆度等即可

经过漫射光的打光后，在表面的瓶口、瓶底及瓶身比较亮，当然回收箱的边缘也比较亮，这里我们只要通过亮度阈值分割；区域面积、宽、高、圆度即可区分出正常瓶子、数量、异物等

阈值分割出背景区域
阈值分割出瓶子区域
作差，优化瓶子区域
依靠宽、高提取大的异物；并做开运算去除箱体边缘等影响
有效区域减去异物
依靠宽、高去除其他影响，并填充剩余区域
依靠圆度分割出瓶子区域
依靠半径检测出较大的圆，即为倒放的瓶子
结果处理

2.1 初步检出瓶子区域

* 分割出背景区域
threshold (Image, BackgroundRegion, 50, 130)
opening_circle (BackgroundRegion, BackgroundRegion, 3.5)

* 分割出有效区域
threshold (Image, Region, 85, 255)

* 作差，优化有效区域
difference (Region, BackgroundRegion, Region)
connection (Region, ConnectedRegions)

blister_04_Pills

2.2 检测较大的异物

* 分割出较大的异物
select_shape (ConnectedRegions, Clutter, ['width','height'], 'or', [100,100], [500,400])
opening_circle (Clutter, Clutter, 8.5)

* 作差减去异物
difference (ConnectedRegions, Clutter, Region)
connection (Region, ConnectedRegions)

在这里插入图片描述

2.3 依靠宽、高去除其他影响，并填充剩余区域

* 有效区域的宽、高范围
select_shape (ConnectedRegions, Candidates, ['width','height'], 'and', [25,25], [100,100])

* 填充区域
fill_up (Candidates, FilledCandidates)

* 使用开运算去除小区域
opening_circle (FilledCandidates, FilledCandidates, 15.5)

在这里插入图片描述

2.4 依靠圆度分割出瓶子区域

* 圆度分割瓶子区域
select_shape (FilledCandidates, RoundCandidates, 'circularity', 'and', 0.87, 1)

* 依靠半径筛选大区域
select_shape (RoundCandidates, BigBottles, 'max_diameter', 'and', 75, 99999)

在这里插入图片描述

3. 全部代码

* This example counts bottles in bottle crates of returned empties.
* With diffuse front light, correctly inserted bottles can
* be segmented relatively easy with a simple thresholding operation,
* followed by some basic morphology operations.
* Bottles inserted upside-down are also detected and displayed
* in orange, while across lying bottles or big objects are
* marked as clutter and produce a warning.
* 
* init
get_system ('store_empty_region', StoreEmptyRegion)
set_system ('store_empty_region', 'false')
read_image (Image, 'bottles/bottle_crate_01')
dev_close_window ()
dev_open_window_fit_image (Image, 0, 0, -1, -1, WindowHandle)
set_display_font (WindowHandle, 16, 'mono', 'true', 'false')
dev_update_off ()
dev_set_draw ('margin')
* main loop
for Index := 1 to 24 by 1
    read_image (Image, 'bottles/bottle_crate_' + Index$'.02')
    * extract background
    threshold (Image, BackgroundRegion, 50, 130)
    opening_circle (BackgroundRegion, BackgroundRegion, 3.5)
    * extract bottle reflections
    threshold (Image, Region, 85, 255)
    difference (Region, BackgroundRegion, Region)
    connection (Region, ConnectedRegions)
    * large reflections are judged as clutter
    select_shape (ConnectedRegions, Clutter, ['width','height'], 'or', [100,100], [500,400])
    opening_circle (Clutter, Clutter, 8.5)
    difference (ConnectedRegions, Clutter, Region)
    connection (Region, ConnectedRegions)
    * ignore noise regions
    select_shape (ConnectedRegions, Candidates, ['width','height'], 'and', [25,25], [100,100])
    fill_up (Candidates, FilledCandidates)
    opening_circle (FilledCandidates, FilledCandidates, 15.5)
    * only 'round' reflections are from bottles
    select_shape (FilledCandidates, RoundCandidates, 'circularity', 'and', 0.87, 1)
    select_shape (RoundCandidates, BigBottles, 'max_diameter', 'and', 75, 99999)
    * display results
    dev_set_shape ('ellipse')
    count_obj (RoundCandidates, NumBottles)
    count_obj (Clutter, NumClutter)
    dev_display (Image)
    dev_set_line_width (5)
    dev_set_color ('lime green')
    dev_display (RoundCandidates)
    dev_set_color ('goldenrod')
    dev_display (BigBottles)
    dev_set_color ('red')
    dev_display (Clutter)
    dev_set_shape ('original')
    disp_message (WindowHandle, NumBottles + ' bottles found.', 'window', 12, 12, 'black', 'true')
    if (NumClutter > 0)
        disp_message (WindowHandle, 'Warning! Clutter detected!', 'window', 40, 12, 'red', 'true')
    endif
    if (Index < 24)
        disp_continue_message (WindowHandle, 'black', 'true')
        stop ()
    endif
endfor
* restore original system settings
set_system ('store_empty_region', StoreEmptyRegion)