多语言交流背景下R-C3D优化手语识别及采集系统分析

Analysis of R-C3D optimized sign language recognition and acquisition system in the context of multilingual communication

手势是一种自然而直观的人际交流模式,手势识别是实现新一代人机交互所不可缺少的一项关键技术。由于手势本身具有多样性、多义性,以及时间和空间上的差异性等特点,因此,手语识别也是当前研究的热点。研究基于静态手语识别,采用卷积神经网络对手势进行特征信息提取,训练相应网络模型。然后将结果输入到区域卷积三维网络中进行动态连续手语识别。研究表明,使用深度网络特征提取网络后,比较原始区域卷积三维网络,对于连续手语识别效果更优,损失曲线收敛更快,最终损失值更低。改进模型在进行连续手语动作识别时,全类平均精度提高4.59%,速度提高58.16%,识别能力和识别精度大大提高。可见研究算法在手语识别和图像动态识别上具有重要意义。

Gesture is a natural and intuitive mode of interpersonal communication,and gesture recognition is an essential key technology for achieving the new generation of human-computer interaction.Due to the diversity,polysemy,and temporal and spatial differences of gestures themselves,sign language recognition is also a current research hotspot.Research is based on static sign language recognition,using convolutional neural networks to extract feature information from gestures,and training corresponding network models.Then input the results into the regional convolutional 3D network for dynamic continuous sign language recognition.Research has shown that after using deep network feature extraction networks,compared to the original region convolutional 3D network,the performance of continuous sign language recognition is better,the loss curve converges faster,and the final loss value is lower.When improving the model for continuous sign language action recognition,the average accuracy of the entire class increased by 4.59%,the speed increased by 58.16%,and the recognition ability and accuracy were greatly improved.It can be seen that the algorithms studied have significant importance in sign language recognition and image dynamic recognition.