基于深度学习的实时姿态识别算法生成人物二维动画

Real Time Posture Recognition Algorithm Based on Deep Learning to Generate 2D Animation of Characters

目前用于人物二维动画建模的人物成像算法存在精准度不高、执行效率差的问题,还有较大提升空间。基于此,提出了一种基于深度学习的人物二维动画生成算法,对人体运动状态下的姿态进行识别、模拟,对视频图像预处理、人物边缘轮廓绘制提炼,以神经网络结构算法为基准,采用聚类算法对关键关节点数据进行处理,降低数据处理量,提高执行效率。结合人物动作姿态识别和二维动画建模,实现了针对动画人物构建改进的卷积神经网络架构,并运用大数据分析建模最终生成人体二维动画模型。对比经典的算法,明显提高了动作识别的精准度以及算法的执行速度。

At present,the character imaging algorithm used for 2D animation modeling of a character has low accuracy and poor execution efficiency,which have much room for improvement.Thus,a core algorithm based on deep learning is proposed to recognize and simulate people’s postures when they move.Firstly,video image is preprocessed and characters’edge contours are drawn and refined.Based on the neural network structure algorithm,the key joint point data are processed by clustering algorithm to reduce the amount of data processing and improve the execution efficiency.Combined with the recognition of character motion and posture as well as 2D animation modeling,an improved convolution neural network architecture for animated characters is realized,and the modeling is analyzed by big data to finally generate the 2D animation modeling of human body.Compared with the classical algorithm,it significantly improves the accuracy of motion recognition and the execution speed of the algorithm.