虚拟现实中基于轻量级神经网络的手势识别

Gesture recognition based on lightweight neural network in virtual reality

手势是虚拟现实环境中的重要输入方式,手势识别的精准度和效率为虚拟现实环境的有效交互提供了基础。为了提高虚拟现实环境中手势识别的正确率和速度,提出了一种基于轻量级优化卷积神经网络的手势识别模型。首先,对手势图像样本进行特征提取及初始化,并构建卷积神经网络手势识别模型。然后,借助MobileNet V2进行轻量化处理,通过深度与逐点卷积,有效降低手势识别模型参数量和计算量。最后,根据最小化手势识别误差反向求解神经网络的各个参数。试验结果表明,所提轻量级神经网络的手势识别准确率相对于原始卷积神经网络有微小下降,但其识别效率却大幅提升,特别是在大规模手势样本的识别场景下。轻量级卷积神经网络表现出了较高的识别综合性能,在虚拟现实环境下的手势识别适应度较高。

Gesture is an important input mode in virtual reality environment,and the accuracy and efficiency of gesture recognition provide the basis for effective interaction in virtual reality environment.In order to improve the accuracy and speed of gesture recognition in virtual reality environment,a gesture recognition model based on lightweight optimized convolutional neural network is proposed.Firstly,feature extraction and initialization are carried out on gesture image samples,and a convolution neural network gesture recognition model is constructed.Then,with the help of MobileNet V2,the gesture recognition model parameter quantities and calculation amounts are effectively reduced through deep and point-by-point convolution.Finally,the parameters of the neural network are solved by minimizing the gesture recognition error.The experimental results show that the gesture recognition accuracy of the proposed lightweight neural network is slightly lower than that of the original convolutional neural network,but its recognition efficiency is greatly improved,especially in the scene of large-scale gesture sample recognition.The lightweight convolutional neural network shows high recognition comprehensive performance,and the gesture recognition adaptability is high in virtual reality environment.