The main method for communication between deaf-mute people and others is sign language.A neural net-work-based sign language recognition system is designed based on STM32,and the TensorFlow and Keras machine learn-ing frameworks are used to train the network model.Due to the limitations of the training data,L2 regularization is intro-duced to solve the overfitting problem during training.Use the STM32Cube.AI toolkit launched by STMicroelectronics to de-ploy a neural network on the STM32 embedded platform,use the OV2640 camera to collect images and perform zoom processing,and transfer the image data to the network for inference to obtain recognition results.After testing,the network model has achieved good robustness and adaptability.The average inference time per frame is 15.8 ms when running on STM32,and the average accuracy rate is 92.3％.Using STM32 as a sign language recognition terminal greatly improves the portability of the device,saves costs,and responds quickly.It provides a new solution for the communication between deaf-mute and normal people,which has high practical application value.
关键词
手语识别/X-CUBE-AI/过拟合/神经网络/L2正则化/STM32/图像识别
Key words
sign language recognition/X-CUBE-AI/overfitting/neural network/L2 regularization/STM32/image recognition
维普
万方数据