Human Tumble Action Recognition Using Spiking Neuron Network

被引：0

作者：

Li, Yu ^{[1
]}

Wang, Ke ^{[1
]}

Huang, MinFeng ^{[2
]}

Li, RuiFeng ^{[1
]}

Gao, TianZe ^{[1
]}

Wu, Jun ^{[1
]}

机构：

[1] Harbin Inst Technol, Harbin 150001, Peoples R China

[2] China Mobile Hangzhou Informat Technol Co Ltd, Hangzhou 311100, Peoples R China

来源：

PROCEEDINGS OF THE 2019 31ST CHINESE CONTROL AND DECISION CONFERENCE (CCDC 2019) | 2019年

关键词：

Human Tumble Action; Spiking Neuron Network; Action Recognition; STDP;

D O I：

10.1109/ccdc.2019.8832749

中图分类号：

TP [自动化技术、计算机技术];

学科分类号：

0812 ;

摘要：

We present a human tumble action recognition method based on Spiking Neuron Network. The approach adopts a method which can convert a skeleton sequence into a 2D image by equating the corresponding X, Y, Z components of the joint coordinates to the R, G, B components of the image pixels. After the conversion, the final images are fed to a recognized Spiking Neuron Network (SNN) model which is based on mechanisms with increased biological plausibility instead of being subsequently converted by a rate-based network. The SNN architecture uses leaky-integrate-and-fire (LIF) neurons, spike-timing-dependent plasticity (STDP), lateral inhibition and intrinsic plasticity. We do not present any labels to the network. Using this unsupervised learning scheme, the architecture achieves 85% accuracy on our dataset. Experimental results demonstrate that our method means a lot especially when using the unsupervised spike-timing-dependent plasticity algorithm.

引用

页码：5309 / 5313

页数：5

共 18 条

[1] Synaptic modifications in cultured hippocampal neurons: Dependence on spike timing, synaptic strength, and postsynaptic cell type [J].