基于语义分割与迁移学习的手势识别

针对复杂场景下深度相机环境要求高,可穿戴设备不自然,基于深度学习模型数据集样本少导致识别能力、鲁棒性欠佳的问题,提出了一种基于语义分割的深度学习模型进行手势分割结合迁移学习的神经网络识别的手势识别方法。通过对采集到的图像数据集首进行不同角度旋转,翻转等操作进行数据集样本增强,训练分割模型进行手势区域的分割,通过迁移学习卷积神经网络更好的提取手势特征向量,通过Softmax函数进行手势分类识别。通过4个人在不同背景下做的10个手势,实验结果表明: 针对复杂背景环境下能够正确的识别手势。

Gesture recognition based on semantic segmentation and transfer learning

Due to the high requirements for the deep camera environment in complex scenes, wearable devices are not natural, and the lack of data set samples based on the deep learning model leads to poor recognition ability and robustness, A gesture recognition method based on deep learning model based on semantic segmentation and neural network based on transfer learning is proposed. By rotating and flipping the collected image data set at different angles, data set samples were enhanced, segmentation model was trained to segment gesture areas, and gesture feature vectors were extracted better through transfer learning convolutional neural network. Softmax function is used for gesture classification and recognition. Through 10 gestures made by 4 people in different backgrounds, the experimental results show that they can correctly recognize gestures in complex environments.