activity-recognition

The paper proposes a method using motion and tactile features to automatically segment atomic actions from teleoperated demonstrations for complex robotic tasks, provides a dataset of pick-and-place teleoperation with a dexterous hand, and shows the proposed features generalize between episodes and similar objects while tactile sensing improves activity recognition within demonstrations.

Modern service robots are provided with one or more sensors, often including RGB-D cameras, to perceive objects and humans in the …

Dataset is composed of 20 long RGB-D videos. Each video provides RGB-D images and tracked skeleton of different social and individual activities performed by two people.

Dataset of 10 episodes containing a sequence of individual/social interactions in an indoor environment.

Experiments for the ECAI 2016 publication. The experiments show different approaches for learning spatio-temporal context for indoor activity recognition.

This is a social interaction dataset between two subjects. This dataset consists of RGB and depth images, and tracked skeleton data (i.e. joints 3D coordinates and rotations) acquired by an RGB-D sensor.

We investigate how incremental learning of long-term human activity patterns improves the accuracy of activity classification over …

The paper presents a probabilistic approach using spatio-temporal and social features to recognize human-human interaction activities, learns priors based on proxemics theory to improve classification, and provides a new public RGB-D dataset of social activities including risk situations, with results showing the proposed method merging different features and proximity priors improves precision, recall and accuracy over alternative approaches