Recognizing Complex Activities by an Interval-Based Bayesian Generative Model
Li Liu · Yongzhong Yang · Lakshmi Narasimhan · Shu Wang · Yongpo Jia · Jun Zhong · Li Cheng · David S. Rosenblum
We propose a new complex activity dataset from depth imaging consisting of hand actions based on the American sign language (ASL). It is an ongoing effort, and at the moment our dataset is more than 5-fold larger compared to the existing ones. We have ensured that the data has been diversified with respect to gender, age and race. For detailed definitions of the complex activities, atomic activites, data acquisition parameters etc. please refer to the About section. The dataset (with annotations) and the code (for I/O and further processing) can be downloaded from the corresponding links on this page. Key statistics pertaining to the data are highlighted below.
Number of subjects
Hours of recorded video
Number of complex activities
Complex activity categories
Atomic Transitions
Number of frames