Action Recognition Using Multi-layer Topographic Independent Component Analysis

摘要

Action recognition plays an important role in the field such as video supervision and medical diagnosis. Common approaches focused on the extension from two-dimensional hand-designed features to video data, or extracting spatio-temporal features via trajectory. This paper proposes a multilayer neural network to extract invariant spatio-temporal features from large amount of videos based on deep learning methods. First, we use Topographic Independent Component Analysis (TICA) to build a two layer stacked convolutional neural network, obtaining weights from training database. Spatio-temporal features are then quantized into visual words with K-means clustering. Non-linear Support Vector Machine (SVM) is used to classify frequency histogram of visual words into different action groups. We applied our algorithm to Hollywood2 database, extracting spatio-temporal features from 12 human action groups. Results showed that feature weights trained by TICA network were similar with Gabor filter, which have obvious selectivity of frequency and direction, robustness to phase variation, conforming to human visual system.