基于NAO机器人的数字识别

Digital recognition based on NAO robot

机器人拥有多种应用,其中较为重要的是图像识别能力,而它们的图像识别网络均需要不断的训练,才能准确地识别物体。为解决NAO机器人在室内环境下实时数字识别问题,重新设计图像识别模块,在TensorFlow中搭建两种数字识别系统,一种基于BP神经网络,另一种基于卷积神经网络(CNN)。在相同的数据集上,BP神经网络与CNN在仿真中都取得了较好的效果,但在真实的机器人上运行时,CNN在有限的实验次数内得到了更好的数据,被证明是一种更有效的数字识别系统。

Robots have a variety of applications,the most important of which is the image recognition,and the image recognition network requires constant training to accurately identify objects.In order to solve the real⁃time digital recognition problem of NAO robot in indoor environment,the image recognition module was redesigned,and two digital recognition systems were built in the TensorFlow,one based on BP neural network and the other based on CNN(convolutional neural network).The BP neural network and CNN have achieved good results in the simulation on the same data set,but when running on a real robot,CNN gets better data in a limited quantity of experiments,which is proved to be a more efficient digital identification system.