Spatio-temporal feature points detection and extraction based on convolutional neural network
Chaoyu Yang, Qian Liu, Yincheng Liang · Advances in computer science research · 2015
Convolutional neural network is a kind of deep learning model, but it only act on a single image generally.This paper expands the convolutional neural network, studies common spatio-temporal features detection and extraction algorithm, proposes a model for detecting and extracting spatio-temporal features based on convolutional neural network, applicates convolutional neural network in action recognition.This model use a plurality of consecutive video frames as input, extract image feature and time dimension information from sequent video frames, convolute and sub sampling alternately, extract a variety of advanced complex abstract features gradually.Experiments show that spatio-temporal convolution neural network has improved ability to classifing and learning.