Video analysis using a deep fusion reasoning engine (DFRE)

1. A method comprising:

receiving, at a video analysis service, video data captured by one or more cameras at a particular location;

applying, by the service, a neural network-based model to each of a plurality of segmented regions of a frame from the video data, to identify objects within one or more of the plurality of segmented regions of the frame;

mapping, by the service, outputs of the neural network-based model to symbols using a conceptual space that is a geometrical representation defined by one or more quality dimensions, wherein the outputs of the model comprise the identified objects;

applying, by the service, a symbolic reasoning engine to the symbols, to generate an alert; and

sending, by the service, the alert to a user interface in conjunction with the video data.

2. The method as in claim 1 , wherein sending the alert to the user interface in conjunction with the video data comprises:

providing the alert as an overlay for one or more frames of the video data.

3. The method as in claim 1 , wherein applying the neural network-based model comprises:

dividing the frame from the video data into the plurality of segmented regions by applying a segmented Bezier curve approximation to the frame.

4. The method as in claim 1 , wherein applying the neural network-based model comprises:

tracking movement of an object over time across frames from the video data.

5. The method as in claim 1 , wherein mapping outputs of the neural network-based model to symbols using a conceptual space comprises:

applying a seed ontology to the outputs of the neural network-based model.

6. The method as in claim 5 , further comprising:

using a sensori-motor control system to expand the ontology for a particular object.

7. The method as in claim 1 , wherein the identified objects comprise a shelf and one or more items on the shelf.

8. The method as in claim 7 , wherein the alert is indicative of an item availability on the shelf.

9. The method as in claim 1 , wherein the identified objects comprise a vehicle and a pedestrian.

10. An apparatus, comprising:

one or more network interfaces to communicate with a network;

a processor coupled to the network interfaces and configured to execute one or more processes; and

a memory configured to store a process executable by the processor, the process when executed configured to:

receive video data captured by one or more cameras at a particular location;

apply a neural network-based model to each of a plurality of segmented regions of a frame from the video data, to identify objects within one or more of the plurality of segmented regions of the frame;

map outputs of the neural network-based model to symbols using a conceptual space that is a geometrical representation defined by one or more quality dimensions, wherein the outputs of the model comprise the identified objects;

apply a symbolic reasoning engine to the symbols, to generate an alert; and

send the alert to a user interface in conjunction with the video data.

11. The apparatus as in claim 10 , wherein the apparatus sends the alert to the user interface in conjunction with the video data by:

providing the alert as an overlay for one or more frames of the video data.

12. The apparatus as in claim 10 , wherein the apparatus applies the neural network-based model by:

dividing the frame from the video data into the plurality of segmented regions by applying a segmented Bezier curve approximation to the frame.

13. The apparatus as in claim 10 , wherein the apparatus applies the neural network-based model by:

tracking movement of an object over time across frames from the video data.

14. The apparatus as in claim 10 , wherein the apparatus maps outputs of the neural network-based model to symbols using a conceptual space comprises:

applying a seed ontology to the outputs of the neural network-based model.

15. The apparatus as in claim 14 , wherein the process when executed is further configured to:

use a sensori-motor control system to expand the ontology for a particular object.

16. The apparatus as in claim 10 , wherein the identified objects comprise a shelf and one or more items on the shelf.

17. The apparatus as in claim 16 , wherein the alert is indicative of an item availability on the shelf.

18. The apparatus as in claim 10 , wherein the identified objects comprise a vehicle and a pedestrian.

19. A tangible, non-transitory, computer-readable medium storing program instructions that cause a video analysis service to execute a process comprising:

receiving, at the video analysis service, video data captured by one or more cameras at a particular location;

applying, by the service, a neural network-based model to each of a plurality of segmented regions of a frame from the video data, to identify objects within one or more of the plurality of segmented regions of the frame;

mapping, by the service, outputs of the neural network-based model to symbols using a conceptual space that is a geometrical representation defined by one or more quality dimensions, wherein the outputs of the model comprise the identified objects;

applying, by the service, a symbolic reasoning engine to the symbols, to generate an alert; and

sending, by the service, the alert to a user interface in conjunction with the video data.

20. The computer-readable medium as in claim 19 , wherein sending the alert to the user interface in conjunction with the video data comprises:

providing the alert as an overlay for one or more frames of the video data.