Recognizing minutes-long activities in videos

An apparatus for classifying subject activities in videos, the apparatus comprising:

• a memory;
• and
• at least one processor coupled to the memory, the at least one processor configured:
• to receive video segments of a video;
• to measure a similarity between the video segments and previously generated concepts to obtain segment representations as a weighted set of previously generated concepts;
• to determine a relationship between the segment representations and their transitioning pattern over time to determine a reduced set of nodes and/or edges for a graph;
• to generate the graph of an activity in the video represented by the video segments based on the reduced set of nodes and/or edges, where nodes of the graph are represented by the previously generated concepts;
• and
• to classify subject activities in the video based on the graph.

A non-transitory computer-readable medium having program code recorded thereon for classifying subject activities in videos, the program code executed by a processor and comprising:

• program code to receive video segments of a video;
• program code to measure a similarity between the video segments and previously generated concepts to obtain segment representations as a weighted set of previously generated concepts;
• program code to determine a relationship between the segment representations and their transitioning pattern over time to determine a reduced set of nodes and/or edges for a graph;
• program code to generate the graph of an activity in the video represented by the video segments based on the reduced set of nodes and/or edges, where nodes of the graph are represented by the previously generated concepts;
• and
• program code to classify subject activities in the video based on the graph.

An apparatus for classifying subject activities in videos, comprising:

• means for receiving video segments of a video;
• means for measuring a similarity between the video segments and previously generated concepts to obtain segment representations as a weighted set of previously generated concepts;
• means for determining a relationship between the segment representations and their transitioning pattern over time to determine a reduced set of nodes and/or edges for a graph;
• means for generating the graph of an activity in the video represented by the video segments based on the reduced set of nodes and/or edges, where nodes of the graph are represented by the previously generated concepts;
• and
• means for classifying subject activities in the video based on the graph.