Cycle detection techniques
Techniques for detecting cycle data can include determining object properties and motion properties in a set of consecutive frames of a sensor stream. The cycle data can be determined from the object properties and motion properties without detecting constituent objects. The object properties and motion properties enable improved detection of cycle data in the presence of different object poses, different positions of the object, partial occlusion of the object, varying illumination, variation in the background, and or the like.
1. A cycle detection method comprising:
receiving a sensor stream including constituent objects of a plurality of cycles of a process;
determining object properties and motion properties of the constituent objects in sets of consecutive frames of the sensor stream, including
generating a set of frame descriptor vectors for a corresponding set of sequential image frames;
concatenating the set of frame descriptor vectors to generate a joint descriptor vector;
generating a fully convolved descriptor vector from the joint descriptor vector;
generating probabilities of one or more of a cycle-start event, a cycle-continue event, a cycle-end event and a no-cycle event based on the fully convolved descriptor vector;
labeling a plurality of grid points with a maximum probability of one or more of the cycle-start event, a cycle-continue event, a cycle-end event and a no-cycle event probabilities; and
performing a connected component analysis to determine cycle probabilities for one or more of the cycle-start event, a cycle-continue event, a cycle-end event and a no-cycle event; and
determining cycle-start events, cycle-continued events, cycle-end events and no-cycle events of the plurality of cycles of the process based on the determined object properties and motion properties of the constituent objects.
2. The cycle detection method of claim 1 , wherein determining the cycle-start event comprises determining an absence of the object in a previous frame and a current frame and a presence of the object in a next frame.
3. The cycle detection method of claim 1 , wherein determining the cycle-end event comprises determining a presence of the object in a previous frame and a current frame and an absence of the object in a next frame.
4. The cycle detection method of claim 1 , further comprising: determining cycle data based on one or more of the determined cycle-start events, cycle-continue events, cycle-end events and no-cycle events.
5. The cycle detection method of claim 4 , further comprising:
determining statistical data based on the determined cycle data;
storing the sensor stream and one or more of the determined cycle-start events, cycle-continue events, cycle-end events and no-cycle events, the determined cycle data, and the determined statistical data indexed to corresponding portions of the sensor stream.
6. The cycle detection method of claim 1 , wherein there are a plurality of different poses of the object in a plurality of different frames of the received sensor stream.
7. The cycle detection method of claim 1 , wherein there are a plurality of different positions of the object in a plurality of different frames of the received sensor stream.
8. The cycle detection method of claim 1 , wherein there is one or more occlusions of the object in a plurality of different frames of the received sensor stream.
9. The cycle detection method of claim 1 , wherein there are illumination differences between a plurality of different frames of the received sensor stream.
10. The cycle detection method of claim 1 , wherein there are difference in a background between a plurality of different frames of the received sensor stream.