Realtime Multi-Person Pose Estimation Based on Android System
Weijian Chen, Shujing Wang, Jingeng Wang · 2020
Human pose estimation has always been a very popular direction. This technology can be applied to human-computer interaction, abnormal behavior detection, intelligent security and other fields, and it will promote the development of artificial intelligence in the future. Human pose estimation has been developed for a long time, and with the advancement of science and technology, many emerging algorithms have spewed out, but most of these algorithms use a multistage network structure. These network structures have achieved better results in accuracy, but the parameters have become larger and larger. It's a big problem that deploying these networks to mobile and embedded devices with limited system resources. We use the MobileNet and TensorFlow Lite frameworks launched by Google to successfully optimize the complex network model and deploy it to Android phones. The human pose estimation can be quickly realized only by using local calculations, avoiding the delay trouble caused by cloud computing solutions.