基于C64xDSP的智能相机技术及应用

发布时间：2018-03-28 00:28

本文选题：DM642　切入点：多处理器　出处：《天津大学》2012年硕士论文

【摘要】：生产流水线的出现使得生产制造的效率有了超乎想象的提高,高速的生产要求检测速度相应加快,这给传统的检测方式提出巨大的挑战。利用智能相机实现视觉检测的方法是一种新兴的检测手段,由于视觉检测设备性能的飞速发展,使得其在流水线上检测的精度、实时性能够得到保证。在实验室已有的DSP5509处理VGA信号的视觉系统基础上,设计了基于C64x DSP核的高速方案。针对目标检测可能遇到的粗大噪声干扰、目标不突出、外部参数易飘变的问题,设计相应的有鲁棒性与自适应性的图像处理算法,提高系统的智能化。完成相机设计后,利用相机设备搭建检测系统,进行了标定,并根据标定结果获得了测量系统的性能参数。利用已标定的系统,设计方案测量了轴承的内外径。文中的具体工作如下: 1.根据轴承检测生产线建立了生产流水线检测的模型,并着重分析了机器视觉在流水线生产中的动态定位模型。推导得出了流水线上目标定位的递推公式。并证明该公式有效的提高了流水线目标的定位速度。 2.设计了智能相机的样机。选用MT9M001视觉传感器与原有使用VGA信号的系统相比图像分辨率提高4倍;对FPGA子系统扩展SDRAM,使得前端子系统可以实现暂存、处理;使用了专用的多媒体高速处理芯片DM642,峰值处理能力要比原有5509系统高12倍;并在FPGA与DSP之间设计了乒乓结构与EDMA协同工作的模式,测试数据传输速度由原来的12MByte/s提高到了30MByte/s。 3.对系统进行了有鲁棒性与自适应性的柔性算法设计。讨论了种子生长标记法的改进以提高目标标记速度;提出目标加权匹配法提高目标与背景的显著度;研究神经网络融入系统的方案使智能相机有自学习能力。并做实验验证。 4.自主设计了利用轴承端面影像测量轴承直径的方法。提出方案,先采样拟合求得中心位置,再以此中心做极坐标转换求解的内外径检测方案对轴承目标进行检测。使检测在保证鲁棒性的基础上有较高精度。
[Abstract]:The emergence of production pipeline has increased the efficiency of production and manufacturing beyond imagination, and the high speed of production requires that the speed of detection be quickened accordingly. This brings a great challenge to the traditional detection methods. The method of using intelligent camera to realize visual detection is a new detection method. Because of the rapid development of the performance of the visual inspection equipment, it makes the accuracy of the inspection on pipeline. The real-time performance can be guaranteed. Based on the visual system of DSP5509 processing VGA signal in laboratory, a high speed scheme based on C64x DSP core is designed. In order to improve the intelligence of the system, the corresponding image processing algorithm with robustness and adaptability is designed to improve the intelligence of the system. After the camera design is completed, the detection system is built and calibrated by using the camera equipment. According to the calibration results, the performance parameters of the measuring system are obtained, and the inner and outer diameter of the bearing is measured by using the calibrated system. The specific work in this paper is as follows:. 1. According to the production line of bearing testing, the model of production pipeline detection is established. The dynamic positioning model of machine vision in pipeline's production is analyzed emphatically, and the recursive formula of the target location on pipeline is deduced, and it is proved that the formula can effectively improve the speed of the target location on pipeline. 2. The prototype of the intelligent camera is designed. Compared with the original system using VGA signal, the image resolution of the MT9M001 vision sensor is increased by 4 times, and the FPGA subsystem is extended to make the front-end subsystem can be stored and processed temporarily. The special multimedia high speed processing chip DM642 is used, the peak processing capacity is 12 times higher than the original 5509 system, and the ping-pong structure is designed between FPGA and DSP to work together with EDMA. The speed of data transmission is improved from the original 12MByte/s to 30 MByte / s. 3. The flexible algorithm with robustness and adaptability is designed, the improvement of seed growth marker method is discussed to improve the target marking speed, and the target weighted matching method is proposed to improve the saliency of target and background. This paper studies the scheme of neural network integration into the system and makes the intelligent camera have the ability of self-learning. 4. The method of measuring the diameter of the bearing by using the image of the end face of the bearing is designed. The scheme is put forward, and the center position is obtained by sampling and fitting. Then the inner and outer diameter detection scheme based on polar coordinate transformation is used to detect the bearing target, which makes the detection have a high accuracy on the basis of the robustness.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP368.1;TH74

【相似文献】