Research And Development Of Embedded Machine Vision Control System Based On Cortex-A8

Today in the industrial production, machine vision applications are more common, and the performance of machine vision controller become more powerful with the continuous development of electronic technology. Among them, the embedded machine vision system has been widely used in detection of the industry because of its small size, low power consumption and adaptability etc.This paper has analyzed the research trends of embedded machine vision systems, and identified hardware and software solutions based on Cortex-A8 AM3354 processor with considering the performance requirements of industrial inspection. The FPGA drives the CMOS image sensor, AM3354 controls the CMOS sensor to capture images with FPGA. Carrying the μC/OS-III real-time operating system build a multitasking platform, and porting Lw IP Ethernet protocol achieve reliable communication of a large of visual information.In the ethernet communication of vision control system, when a large of image data is transmitting, the data transmission efficiency is low and there is spurious retransmission because RTO lag the change of RTT. This paper proposes a modified network congestion prevention method of Lw IP for that. In order to overcome the transmission efficiency effect of repeated rapid recovery or slow-start caused by multiple packet loss in a send window, this paper has proposed confirming the serial mechanism in the TCP congestion control mechanisms to determine data is received completely or not by the judge of the serial numbers. In order to improve the inaccurate of RTO estimate and reduce TCP data transmission spurious retransmission, the normalized least mean square error algorithm can estimate the value of RTT and calculate the RTO value more accurately. Experimental results show that the improved congestion control method can more effectively deal with packet loss, effectively reduce spurious retransmission of data, and improve the data throughput.Due to the increasing complexity of image detection algorithm, real-time requirements of embedded systems become more stringent. Based on high requirements of the industrial field, this paper achieves parallel to optimize the significant areas detection of image. By analyzing the structure of the algorithm, in the detection process, opening successive storage area describe the branch regions and convenient the parallel computing regional variation rate and the search of maximally stable extremal regions. According to the characteristics of the algorithm calculation module, parallel processing architecture is designed. And it has optimized the calculate of extreme value area, geometric moments of region and the central matrix, and accelerate the maximally stable extremal regions detection and the ellipse fitting of salient feature regions. Experimental results show that optimization of image saliency region detection algorithm can extract the salient region features more quickly and effectively.In order to verify the overall performance of the proposed embedded machine vision control system, it has constructed multitasking execution platform. The system display the acquired image and image after significant image region detection on PC through ethernet communication, and the test show that the system is running well.