如何克服复杂生产环境对视觉检测系统数据准确性的干扰
发布时间：
2023-02-10 11:28
当面对复杂的检测物体时，机器视觉检测系统是否能够出色的完成检测任务呢？
When faced with complex detection objects, can machine vision detection system perform the detection task excellently?
工业自动化生产环境中，机器视觉检测系统的主要功能分别是识别计数、视觉定位、尺寸测量和外观检测，但是面对生产环境中的复杂性，环境光、震动、烟尘等因素会对干扰视觉检测系统对3D信息获取的准确性，而克服这一问题的关键就是就是应用多种方法增强3D信息获取时的环境适应性。
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.
在外观缺陷检测上，机器视觉通常会使用3D成像，而当前3D机器视觉研究的核心在于如何实现成像过程的可逆，即如何由2D信息恢复成3D信息，其中最为关键的点在于如何获取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.
投影正弦结构光增强工业场景的纹理并加速场景3D视觉信息的提取。利用正弦结构光投影的方法，可以实现亚像素的点云计算，能够保证工业工件的识别和精确定位定姿。
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.
我们对利用结构光进行场景3D信息获取的点云计算方法实现了优化，可以实现1秒内完成拍摄与点云计算。从而实现了工业场景的快速建模，达到工业生产节奏的需求。
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.
当面对目前复杂被测物时，厂商如果需要通过机器视觉来进行3D图像的检测，通常办法为加强工业场景的纹理，使用结构光向物体进行投影，再通过点云计算来根据投影变形的情况计算出被测物体表面的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.
在具体应用场景中，使用机器视觉检测系统进行产品检测时，生产线中的产品通常并不会停下来等待检测，而会以一种匀速通过，排除一些通过抽检方式进行检测的情况。如果想要不影响生产效率，只能在生产线中进行动态检测，而这对于机器视觉检测系统的计算要求也将变得极高。
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.
现有的3D视觉检测系统方案对于漫反射材质的物体能够实现良好的数据采集，对于如透明或者高反光的物体表现不佳，在这一块知微传感在研的一款新产品可以很好的解决高反物体在不同角度下的数据采集，将采用多目动态结构光的方案，同时产品也将继承RGB摄像头，可以使用RGB信息通过深度学习的方法实现复杂物品的分割，更好适应产线的自动化。
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.
视觉检测系统