Human motion recognition (HMR) is playing an increasingly key role in many fields including public security, medical treatment and health care. In this paper, we propose a fractional Fourier transform (FrFT) based cadence-velocity diagram (CVD) based method, to improve the classification rate, which can effectively distinguish similar human motions in certain traditional feature domains such as the time-frequency (TF) domain. Besides, we also incorporate the feature in FrFT based CVD domain with the range feature, which can be regarded as the multi-domain feature. Then six human daily motions are then classified by the convolutional neural network (CNN) with the above multi-domain feature. Experimental results based on real data has demonstrated that the proposed method can achieve a high classification rate.