实时FPGA手势识别算法的设计

Design of real-time figure recognition algorithm based on FPGA

针对手势识别的硬件加速问题,对图像识别技术进行研究,从手势边缘与肤色信息出发,借助FPGA平台的高速特性,在Sobel边缘提取算法的基础上,提出利用椭圆颜色域分离法与高斯函数加权来优化手势信息特征值提取的算法。原始图像分别输入Sobel边缘提取模块与色彩空间分离模块,经处理后的信号共同输入高斯加权模块,与肤色相关的边缘信息得到加强,无关信息得到抑制。该算法结合了手势图像的色彩信息与边缘信息,经过对比验证,表明通过对边缘与肤色信息的算法优化,能有效滤去与手势无关的信息,在不同光照下测试均获得足够的辨识度。

To implement hardware acceleration on the recognizing of figures,the optimization algorithm based on the high-speed performance of distributed field-programmable gate arrays(FPGA)platform and the study of the hand edge and skin color information is proposed.It uses Sobel edge detection algorithm to extract contour information and adopts the ellipse color space division method and Gauss function weighting structure to filter the irrelavance.The original image was input into Sobel edge extraction module and color space separation module respectively,then the processed signals were input into Gaussian weighting module together,so that the edge information related to skin color strengthened and the unrelated information suppressed.The result shows that the algorithm can perfectly realize the combination of skin color and contours and filter the irrelevant information effectively,hence a great improvement to the ability of discerning figures under varies of light conditions.