Abstract: An apparatus includes a processor. The processor receives an algorithmic shopping cart that includes a first set of items determined by an algorithm to have been selected by a person during a shopping session in a physical store, based on a set of inputs received from sensors located within the physical store. The processor also receives a virtual shopping cart that includes a second set of items. Video of the shopping session was captured by a set of cameras located in the physical store and depicts the person selecting the second set of items. The processor compares the algorithmic cart to the virtual cart and determines that a discrepancy exists between the algorithmic cart and the virtual cart. The processor determines a subset of the set of inputs associated with the discrepancy and attaches metadata explaining the discrepancy to the subset. The processor uses the subset to train the algorithm.

Abstract: A system includes sensors and a tracking subsystem. The subsystem receives a first image feed from a first sensor and a second image feed from a second sensor. The field-of view of the second sensor at least partially overlaps with that of the first sensor. The subsystem detects, in a frame from the first feed, a first contour associated with an object. The subsystem determines, based on pixel coordinates of the first contour, a first pixel position of the object. The subsystem detects, in a frame from the second feed, a second contour associated with the same object. The subsystem determines, based on pixel coordinates of the second contour, a second pixel position of the object. Based on the first pixel position and the second pixel position, a global position for the object is determined in a space.

Abstract: A scalable tracking system processes video of a space to track the positions of people within a space. The tracking system determines local coordinates for the people within frames of the video and then assigns these coordinates to time windows based on when the frames were received. The tracking system then combines or clusters certain local coordinates that have been assigned to the same time window to determine a combined coordinate for a person during that time window.

Abstract: An object tracking system includes a sensor and a tracking system. The sensor is configured to capture a first frame of a global plane for at least a portion of a marker grid in a space. The tracking system is configured to receive a first coordinate in the global plane for a first corner of a marker grid, to determine a second coordinate in the global plane for the first marker on the marker grid, and to determine a third coordinate in the global plane where the second marker on the marker grid. The tracking system is further configured to determine a first pixel location for the first marker, to determine a second pixel location for the second marker, and to generate a homography based on the second coordinate for the first marker, the third coordinate for the second marker, the first pixel location, and the second pixel location.

Abstract: A system includes a sensor, a weight sensor, and a tracking subsystem. The tracking subsystem receives an image feed of top-view images generated by the sensor and weight measurements from the weight sensor. The tracking subsystem detects an event associated with an item being removed from a rack in which the weight sensor is installed. The tracking subsystem determines that a first person or a second person may be associated with the event. In response to determining that the first or second person may be associated with the event, buffer frames are stored of top-view images generated by the sensor during a time period associated with the event. The tracking subsystem then determines, using at least one of the stored buffer frames and a first action-detection algorithm, whether an action associated with the event was performed by the first person or the second person.

Abstract: A system includes a sensor, a weight sensor, and a tracking subsystem. The tracking subsystem receives an image feed of top-view images generated by the sensor and weight measurements from the weight sensor. The tracking subsystem detects an event associated with an item being removed from a rack in which the weight sensor is installed. The tracking subsystem determines that a first person and a second person may be associated with the event. The tracking subsystem then determines, using a first approach, whether an action associated with the event was performed by the first person or the second person. If results of the first approach do not satisfy criteria, a second approach is used to assign the action to the first or second person.

Abstract: An object tracking system includes a sensor and a tracking system. The sensor is configured to capture a first frame of a global plane for at least a portion of a marker grid in a space. The tracking system is configured to receive a first coordinate in the global plane for a first corner of a marker grid, to determine a second coordinate in the global plane for the first marker on the marker grid, and to determine a third coordinate in the global plane where the second marker on the marker grid. The tracking system is further configured to determine a first pixel location for the first marker, to determine a second pixel location for the second marker, and to generate a homography based on the second coordinate for the first marker, the third coordinate for the second marker, the first pixel location, and the second pixel location.

Abstract: An apparatus includes a processor. The processor receives an algorithmic shopping cart that includes a first set of items determined by an algorithm to have been selected by a person during a shopping session in a physical store, based on a set of inputs received from sensors located within the physical store. The processor also receives a virtual shopping cart that includes a second set of items. Video of the shopping session was captured by a set of cameras located in the physical store and depicts the person selecting the second set of items. The processor compares the algorithmic cart to the virtual cart and determines that a discrepancy exists between the algorithmic cart and the virtual cart. The processor determines a subset of the set of inputs associated with the discrepancy and attaches metadata explaining the discrepancy to the subset. The processor uses the subset to train the algorithm.