基于神经网络和光流场的嵌入式高速目标识别与跟踪（英文）

Embedded Design in Neural Network and Optical Flow Based High-speed Target Recognition & Tracing

研究了面向嵌入式硬件平台的高速视频图像处理与高速目标跟踪算法。基于神经网络的光流场算法能有效减少程序对硬件系统的内存要求和运算要求,该算法能够比传统算法更加容易的应用于以DSP(数字信号处理)芯片为核心的嵌入式硬件平台中。首先设计了一个基于Hopfield神经网络控制器的自适应滤波器,以图像信噪比为控制指标,对需进行目标识别的视频图像进行图像预处理;然后,利用补偿模糊神经网络控制器对光流场计算方法进行优化,这是一种通过参数控制平滑度实现的平均速度的角度误差和标准角误差的CFNN(补偿模糊神经网络)控制器识别跟踪算法。微机仿真及嵌入式系统试验结果均表明:该算法能够在同等条件下显著提高目标辨别与跟踪能力,显示其具有较高的有效性和实用性。

High-speed video image processing and high-speed target tracing algorithm for embedded hardware platform were analyzed.Optical flow field algorithm based on neural network was effective to reduce program memory and computing resource requirements of hardware system,which can be more suitable for embedded hardware platforms with DSP（digital signal processing）chip as kernel than traditional strategies.Hopfield Neural Network controller was applied to design a self-corrected filter and taking image signal-to-noise ratio as control index,image preprocessing was carried out for target recognition video image.Compensatory fuzzy neural network controller was applied to optimize optical flow field calculation,which was through using parameter of control smoothness to realize the reduction of average velocity angle error standard angular error by CFNN（compensatory fuzzy neural network）controller algorithm.The results show that the algorithm can significantly improve the moving target recognition and tracing ability,which proves that it is a more practical and more effective method.