如何克服復雜生產環境對視覺檢測系統數據準確性的干擾
發布時間：
2023-02-10 11:28
當面對復雜的檢測物體時，機器視覺檢測系統是否能夠出色的完成檢測任務呢？
When faced with complex detection objects, can machine vision detection system perform the detection task excellently?
工業自動化生產環境中，機器視覺檢測系統的主要功能分別是識別計數、視覺定位、尺寸測量和外觀檢測，但是面對生產環境中的復雜性，環境光、震動、煙塵等因素會對干擾視覺檢測系統對3D信息獲取的準確性，而克服這一問題的關鍵就是就是應用多種方法增強3D信息獲取時的環境適應性。
In the industrial automation production environment, the main functions of machine vision inspection system are identification and counting, visual positioning, size measurement and appearance inspection. However, in the face of the complexity of the production environment, environmental light, vibration, smoke and other factors will interfere with the accuracy of the visual inspection system to obtain 3D information, and the key to overcome this problem is to apply a variety of methods to enhance 3D Environmental adaptability of information acquisition.
在外觀缺陷檢測上，機器視覺通常會使用3D成像，而當前3D機器視覺研究的核心在于如何實現成像過程的可逆，即如何由2D信息恢復成3D信息，其中最為關鍵的點在于如何獲取3D信息。
In appearance defect detection, machine vision usually uses 3D imaging, and the core of current 3D machine vision research is how to realize the reversibility of the imaging process, that is, how to recover 2D information into 3D information. The most important point is how to obtain 3D information.
投影正弦結構光增強工業場景的紋理并加速場景3D視覺信息的提取。利用正弦結構光投影的方法，可以實現亞像素的點云計算，能夠保證工業工件的識別和精確定位定姿。
Projecting sinusoidal structured light enhances the texture of industrial scene and accelerates the extraction of 3D visual information. Using the method of sinusoidal structured light projection, the sub-pixel point cloud calculation can be realized, which can ensure the industrial workpiece recognition and accurate positioning and pose determination.
我們對利用結構光進行場景3D信息獲取的點云計算方法實現了優化，可以實現1秒內完成拍攝與點云計算。從而實現了工業場景的快速建模，達到工業生產節奏的需求。
We optimized the point cloud computing method of 3D scene information acquisition using structured light, which can complete shooting and point cloud computing within 1 second. So as to realize the rapid modeling of industrial scene and meet the demand of industrial production rhythm.
當面對目前復雜被測物時，廠商如果需要通過機器視覺來進行3D圖像的檢測，通常辦法為加強工業場景的紋理，使用結構光向物體進行投影，再通過點云計算來根據投影變形的情況計算出被測物體表面的3D信息。
In general, when the measured object is projected into the 3D image to enhance the texture of the measured object, it is usually used to calculate the texture of the measured object according to the 3D image of the industrial scene.
在具體應用場景中，使用機器視覺檢測系統進行產品檢測時，生產線中的產品通常并不會停下來等待檢測，而會以一種勻速通過，排除一些通過抽檢方式進行檢測的情況。如果想要不影響生產效率，只能在生產線中進行動態檢測，而這對于機器視覺檢測系統的計算要求也將變得極高。
In specific application scenarios, when the machine vision inspection system is used for product detection, the products in the production line usually do not stop to wait for detection, but will pass at a uniform speed, excluding some cases of sampling inspection. If you want to not affect the production efficiency, only dynamic detection can be carried out in the production line, and this will make the calculation requirements of machine vision inspection system very high.
現有的3D視覺檢測系統方案對于漫反射材質的物體能夠實現良好的數據采集，對于如透明或者高反光的物體表現不佳，在這一塊知微傳感在研的一款新產品可以很好的解決高反物體在不同角度下的數據采集，將采用多目動態結構光的方案，同時產品也將繼承RGB攝像頭，可以使用RGB信息通過深度學習的方法實現復雜物品的分割，更好適應產線的自動化。
The existing 3D vision detection system can achieve good data acquisition for objects with diffuse reflectance, but not for transparent or highly reflective objects. In this piece of micro sensing, a new product under development, can well solve the data acquisition of highly reflective objects at different angles. The multi camera dynamic structured light scheme will be adopted, and the product will also inherit the RGB camera Firstly, RGB information can be used to realize the segmentation of complex objects by deep learning, which can better adapt to the automation of production line.
視覺檢測系統