| CPC G06T 7/248 (2017.01) [G06T 7/292 (2017.01); G06V 10/32 (2022.01); G06V 20/52 (2022.01); G06T 2207/10016 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/30196 (2013.01); G06T 2207/30232 (2013.01); G06V 2201/10 (2022.01)] | 39 Claims |
|
1. A computer-implementation method for person-tracking across video streams, the method comprising:
retrieving a plurality of video streams generated from a plurality of cameras positioned at a retail venue;
performing a normalization operation over extracted frames of the plurality of video streams to generate normalized frames;
processing the normalized frames by extracting bounding boxes over at least one person identified from the normalized frames and extracting one or more foot locations associated with each of the at least one person to generate feature vectors and detect attributes, wherein the feature vectors are multidimensional vectors representing the bounding boxes generated by an artificial intelligence model trained to minimize a distance between vectors of the same persons;
merging extracted bounding boxes associated with each of the at least one person identified from the normalized frames in time series to generate tracklets using a Hungarian matching algorithm, wherein each tracklet is associated with a person of the at least one person identified from the normalized frames, and wherein the Hungarian matching algorithm uses three factors to generate a cost function comprising Euclidean distance between feature vectors of bounding boxes, intersection over union of bounding boxes, and time distances between the bounding boxes; and
aggregating the tracklets and metadata associated with polygons drawn on camera views defining user defined sections of at least one area in the video streams to generate analytical outputs.
|