Application of video processing card with integrated DSP in machine vision

Machine vision is an optical recognition system that imitates human vision. It uses cameras and computers to capture, analyze and interpret image content, and then make certain decisions. Since machine vision systems can quickly obtain a large amount of information, are easy to process automatically, and are easy to integrate with design information and processing control information, in modern automated production processes, people widely use machine vision systems in fields such as working condition monitoring, finished product inspection and quality control. The characteristics of machine vision systems are to improve the flexibility and automation of production. In some dangerous working environments that are not suitable for manual operations or where artificial vision is difficult to meet the requirements, machine vision is often used to replace artificial vision; at the same time, in large-scale industrial production processes, using artificial vision to check product quality is inefficient and not accurate. Using machine vision detection methods can greatly improve production efficiency and the degree of automation of production. The biggest advantage of vision is that there is no contact with the observed object, so it will not cause any damage to the observer or the observed, and it is very safe and reliable.

1 Composition of machine vision system

Machine vision is generally composed of the following parts: lighting part; optical system, including image sensor and camera; image acquisition card/image digitizer, scanner; visual multiplexer; industrial computer; control actuator. As shown in Figure 1.



2 Problems in traditional machine vision technology

The work in the machine memory is completed by the host CPU. In some cases where the image data volume is large and the algorithm is complex, the CPU of a single machine often cannot complete the calculation in real time. At this time, a local area network needs to be formed by several computers, one of which is used as a server, and the other computers perform image processing. The following is a specific example of the copper foil board surface quality online detection system developed by Hongda Instrument Company. The online detection of the copper foil board surface quality requires non-contact online detection of the copper foil board on the line. The size of the board is 1240mm×1080mm (width×length), and the assembly line speed is 0.5m/s; automatic identification of various defects such as scratches, oxidation points, pad damage, foreign matter, etc.; the detection accuracy is 0.2mm.

Because the project requires high detection accuracy and high assembly line speed, the image processing data volume is huge. Based on the current computer processing speed, a single machine cannot complete the inspection task, so they use a multi-machine parallel processing system based on a local area network. (As shown in Figure 2) The camera connected to each computer is only responsible for inspecting a part of the surface of the steel plate (1/4 of the steel plate surface).



According to the requirements of the system, they adopted a 4+1 solution, that is, 4 client computers connected to 4 CCD cameras complete the real-time acquisition and processing of image data, and transmit the data to a server through the local area network. After the data of all clients are integrated on the server, the inspection results are given.

The defects of this solution for the online inspection system of the surface quality of copper foil are obvious. The system is equipped with a total of 5 computers, which is costly, and the system is relatively complex and the reliability is reduced.

2.1 Improvement plan

By integrating high-performance DSP chips on each image acquisition card to form a high-speed image processing card, the DSP replaces the computer's CPU for image processing. At this time, the DSP is equivalent to the CPU of a computer in the above solution. Further, by inserting 4 such high-speed cards on a computer, the PCI bus controller of the host is responsible for controlling and coordinating the occupancy of the bus by the 4 high-speed image processing cards. At this time, the CPU of the host is equivalent to the server in the above solution. This saves the cost of PC and greatly reduces the cost. Figure 3 shows the hardware schematic diagram of this high-speed image processing card.



2.2 Detailed explanation of the key parts of the hardware circuit

The camera inputs the captured video signal into the ADC and digitizes it. Then the digital video signal is input into the high-speed FIFO. Once the digital video data in the FIFO is almost full, the DSP reads this data into the internal RAM and performs algorithmic processing on the digital video signal. The image processing algorithm detects various defects such as scratches, oxidation points, pad damage, foreign matter, etc., and compares them with the standard. If they exceed the standard, they are recorded. The DSP inputs the final calculation result to the PCI bus controller. The PCI bus controller transfers the calculation result to the host memory in DMA mode. At this time, the CPU of the host is equivalent to the server in the above solution. It only performs statistical analysis on the results processed by the 4 video processing cards with integrated high-speed DSP, and does not perform image algorithmic processing. The following introduces the high-speed image processing card with integrated DSP in combination with specific chips (the chip model is related to the specific application).

The ADC chip can be Philips' SAA7114, which supports NTSC/PAL/SECEM formats, with an A/D conversion accuracy of 9b and a parallel output of 8b video output bandwidth of 27 MHz. One byte is output in parallel for each clock cycle (1/27MHz). In addition to outputting digital pixels, SAA7114 also outputs a clock signal for synchronization. The size of the SAA7114 output image can be controlled by setting the relevant registers of SAA7114.

The synchronous FIFO can be TI's SN74ACT7881, which is 1024b×18 in size. The interface speed between the synchronous FIFO and SAA7114 is 27MHz and the width is 8b. The interface speed between the FIFO and the DSP can be configured to be 81MHz and the width is 16b. When the FIFO is almost full of data, a control signal is sent to the DSP to cause the DSP to generate an interrupt and take away the data in the FIFO.

In order to interface with the PCI bus controller, the DSP needs to support the PCI interface. The DSP can be selected from TI's TMS320C6414. In addition to supporting the PCI interface, the main frequency of this DSP reaches 500MHz, and the pipeline depth is 8 levels, so the peak computing power can reach 4000 MI/s. FIFO can directly interface with the external memory of the DSP. The external memory interface of the DSP transfers data to the memory of the DSP through EDMA (enhanced DMA). The transmission of EDMA is triggered by the synchronization signal of the FIFO. The internal storage structure of 6412 is a modified Harvard structure, that is, it is divided into two levels of storage: L1P/L1D/L2P are 16kB/16kB/1024kB respectively. By setting the relevant registers, L2P can be divided into program memory and data memory of different proportions. The image processing algorithm for processing the surface quality of the copper foil board is placed in L1P and the divided L2P, and the digital image data input through FIFO is placed in L1D and the divided L1P. The compiled binary file of the image processing algorithm is burned into the external FLAH, and the program is mapped into the SDRAM after power-on. FIFO, SDRAM, and FLAH are all connected to the DSP through the DSP's external memory interface local bus (EMIF).

The PCI interface of 6414 complies with the PCI2.2 specification: 32b, 33MHz, and the transmission speed is 132MB. The PCI bus controller can be PCI9054. The DSP transfers the final calculation result to the FIFO of PCI9054. Once 9054 detects that the PCI bus is idle, it sends the data in the FIFO. Since the basic transmission mechanism on the PCI bus is a burst transmission, a burst usually consists of an address cycle and one or more data cycles. This linear burst transmission method can ensure that the bus is constantly fully loaded with data, and can effectively use the bus frequency bandwidth to transmit data, reducing unnecessary addressing operations. PCI's unique synchronous operation function ensures that the microprocessor can operate simultaneously with these bus masters. Once the PCI bus is idle, the high-speed image processing card will immediately transfer data to the memory without waiting for the CPU operation to complete

.

High-speed video processing cards with integrated DSP can be widely used in the technical fields involved in machine vision to reduce costs and increase reliability. In the field of product surface quality monitoring, high-speed video processing cards with integrated DSP can be applied to the surface quality detection of cold-rolled steel plates, galvanized steel plates and other color steel plates in the steel production process; the detection of paper surface (including thickness) quality in the papermaking process; the surface detection of plastic products with high surface quality requirements in the plastic production process; the surface quality detection of devices with high surface quality requirements in the electronic product production process, such as wafer surface quality detection.

In some special working environments, there are stringent requirements on the stability, reliability, safety and volume of the host (such as monitoring in mines and supermarkets). The monitoring system based on PC may not meet the requirements in this case. At this time, an embedded system can be designed: refer to Figure 3 and only a few changes are needed: that is, the end of the high-speed image processing card with integrated DSP is not connected to the PCI bus controller 9054, but to a CPU (such as ARM or PowerPC), on which an operating system such as embedded Linux is transplanted to form an embedded application system to meet special requirements.