Development of real-time high-level integrated vision systems and their application to markerless posture recognition

我们提出了一种受大脑视觉信息处理启发的人脸/手臂姿态识别算法，并开发了一个结合专用集成电路的系统，用于算法的实时运行。Gabor滤波作为大脑主要的视觉信息处理模型之一，提取图像局部空间频率特征，被有效地应用于立体视觉和基于生物运动的运动检测。我们还开发了硬件/软件集成系统，如实时主观轮廓生成系统，使用集成电路，执行像素状态的各向异性扩散。

We proposed a face/arm posture recognition algorithm inspired by the brain visual information processing, and developed a system combining dedicated integrated circuits for real-time operation of the algorithm. As one of the brain primary visual information processing models, Gabor filtering, which extracts features of local image spatial frequencies, was effectively applied to stereo vision and moving detection based on biological motion. We also developed hardware/software integrated systems such as a real-time subjective-contour generation system using an integrated circuit that performs anisotropic diffusion of pixel states.