Development of integrated vision systems for humanoid robots with neural temperature modulation function

本文提出了一种目标检测算法，弥补了人脸识别神经模型弹性图匹配的两个功能缺陷：一个是特征表示的复杂性，一个是Gabor滤波器的计算代价。将目标检测算法在现场可编程门阵列中实现，测试目标检测能力。在神经元温度调制同步转变的理论分析中，我们发现了神经元特有的特性：I类Morris-LeCar神经元的频率随温度的降低而增加，呈现鞍结分叉。

We proposed an object detection algorithm, which made up for two functional deficits in a neural model of face recognition, Elastic Graph Matching : One is the complexity of feature representation while one is the computational cost for Gabor filter. The objection detection algorithm was implemented into FPGA, testing the detection ability. In theoretical analysis of neuronal temperature-modulated synchronization transition, we found the unique neuronal property : frequency increases with temperature reduction in the Morris-Lecar neuron of class-I exhibiting saddle-node bifurcation.