The technology has achieved accuracy and processing performance that allows for integration in online production processes.
Allied Vision Technologies’ Bonito camera with a double Full Camera Link Configuration Interface. Source: Allied Vision Technologies
The measurement technology of today has achieved accuracy and processing performance that allows an integration of acquisition and analysis in online production processes. For this purpose, non-contact methods are almost solely used with need of high requirement on robustness and speed.
When realizing an optical 3-D measurement application by laser triangulation method, the spatial and temporal resolution of the sensor is mainly dependent of the measurement accuracy. The availability of high speed cameras is one of the most important requirements for most of the applications.
Machine vision has reached the limits of its specification for available sensors. Future standards like CoaXPress, Camera Link HS and 10GigE will extend the performance bandwidth significantly and allow new application fields. For the transition period, creativity is needed to offer new speed classes even today.
Pictured is the chart representation of the calculation of metering points from a laser profile by 3-D triangulation measurement with center of gravity method. Source: Allied Vision Technologies
High-Speed Camera: Double Camera Link Bandwidth
With their new high-speed cameras, companies like Allied Vision Technologies (AVT) have implemented an interface that guarantees compatibility to the Machine Vision standard Camera Link, but doubles its bandwidth.
A CMOS high-speed camera is able to acquire images with a sensor resolution of 2320 pixels width and 1726 pixels height at 386 full frames per second. The image data is transferred by four MDR connectors at double Camera Link Full Configuration with 10 taps each. The camera is especially suited for recording of fast movements by its electronic full frame shutter.
A challenge of an online inspection process is that a large amount of image data–1.5 Gigabytes per second–not only has to be acquired, but also processed in real-time. For that purpose, the camera is connected to two Silicon Software microEnable IV VD4-CL frame grabbers. These are programmable image acquisition boards which fulfill the acquisition tasks and additionally the processing of the images on the FPGA hardware.
The implementation of the “DMA900 technology” enables the frame grabber series to transfer up to 900 Megabytes per second via PCIe x4 interface into the RAM of the host PC. The camera uses two data channels when it is operating in high speed mode. Each grabber is responsible for the acquisition of one half of each image. When using the triangulation method for 3-D measurements, the coordinates from laser profiles are calculated column by column, which is not disadvantageous for a use of two frame grabbers.
Laser Triangulation for 3-D Measurement
For the 3-D measurement application, the sensor width is the maximum horizontal resolution in direction of the laser profile. The setting for the sensor height determines the resolution capacity of profile height, which reflects the depth information of the z-axis. The geometric dependency between the device under test, angle of camera and laser, determines the maximum deflection of the laser beam projection. This projection defines the useful vertical range of the sensor. If the maximum available bandwidth is used, a reduction of the sensor height settings will automatically follow in a potentially higher frame rate, which will increase the sampling resolution in scan direction. For example, the settings of the camera with a horizontal resolution of 2320 pixels and a vertical sampling of 128 pixels will result in a scanning performance of about 5200 profiles per second. This temporal resolution determines the horizontal sampling rate together with the speed of the device under test.
Pictured is the connection of the Bonito to two microEnable IV VD4-CL of Silicon Software. Source: Allied Vision Technologies
Peak Detector on the Frame Grabber's Internal FPGA
To determine the best fitting measurement coordinate from the laser line, different methods are implemented on the frame grabber. Compared with simple algorithms like threshold or maximum value method, a center of gravity (“COG”) method and the Peak Detector method are used.1 Both methods achieve a high accuracy even at difficult lighting conditions and surface of material. The Peak Detector method moreover highlights by a higher robustness and lower noise sensitivity. These concerns to sensor noise, laser diffusion as well as surface dependent noise. Hereby this method is algorithmically more stable in difficult materials such as translucent or reflective surfaces. Even in applications with material of the laser color, the measurement results in detecting and assigning metering points are more accurate. By implementing the Peak Detector on the internal FPGA of the frame grabber, the complete processing of metering point calculation is running in real-time. A highly parallel implementation supports the maximum speed of Full Configuration Camera Link cameras at 850 Mbytes per second. At the same time, a maximum, theoretical accuracy of 1/64 pixel is achieved.
The determination of the best fitting coordinate of the measurement point is processed column by column for each image. Reflections and non-relevant areas of the image have to be suppressed. In practice, the Peak Detector algorithm has turned out to be one of the most stable and robust method in various environments and applications. Only 1-D images with information about coordinates as combined result of both sensor parts are transferred to the host PC for further processing.
Pictured is the projection of a laser beam on the device under test and display of the binarized image as base of the metering points calculation. Source: Allied Vision Technologies
A 3-D Object's Cloud of Points
A 3-D object is finally modeled from cloud of points.2 Although all subsequent calculations are processed on the CPU, the same high-speed timing is required. This is realized by additional modules available for merging objects from different POC, matching objects with CAD data, calculating volume data, slicing coordinate or processing data for integrator’s software. V&S
1 The Peak Detector method is attributed to Aqsense, a Spain-based company.
2 Cloud of points is attributed to Aqsense and its measurement software environment, SAL3D.
A CMOS high-speed camera is able to acquire images with a sensor resolution of 2320 pixels width and 1726 pixels height at 386 full frames per second.
The image data is transferred by four MDR connectors at double Camera Link Full Configuration with 10 taps each.
The implementation of the “DMA900 technology” enables the frame grabber series to transfer up to 900 Megabytes per second via PCIe x4 interface into the RAM of the host PC.