A descriptor coping with complicated human activities

Human activity discrimination algorithms are concentrated ordinary to upright poses. The algorithms become unsuitable when persons are non-upright and in complicated poses. To solve this problem, this paper proposes a pose encoding descriptor - AMOP to describe human pose precisely. This descriptor is invariant to scale and direction, furthermore it has the feature of easy to compress, expandable and convenient to compute. Experiments verify the convenience and effectiveness of this descriptor.