3D Action Recognition in an Industrial Environment

In this study we introduce a method for 3D trajectory based recognition of and discrimination between different working actions. The 3D pose of the human hand-forearm limb is tracked over time with a two-hypothesis tracking framework based on the Shape Flow algorithm. A sequence of working actions is recognised with a particle filter based non-stationary Hidden Markov Model framework, relying on the spatial context and a classification of the observed 3D trajectories using the Levenshtein Distance on Trajectories as a measure for the similarity between the observed trajectories and a set of reference trajectories. An experimental evaluation is performed on 20 real-world test sequences acquired from different viewpoints in an industrial working environment. The action-specific recognition rates of our system correspond to more than 90%. The actions are recognised with a delay of typically some tenths of a second. Our system is able to detect disturbances, i.e. interruptions of the sequence of working actions, by entering a safety mode, and it returns to the regular mode as soon as the working actions continue.