| CPC G06V 20/52 (2022.01) [G06N 3/08 (2013.01); G06N 20/00 (2019.01); G06V 40/10 (2022.01); G06V 40/171 (2022.01); G06V 20/625 (2022.01); G06V 2201/08 (2022.01)] | 17 Claims |
|
1. A computer-implemented method, comprising:
detecting one or more vehicle objects and one or more human objects in an image, using a single detection function comprising a single artificial intelligence (AI) model, each of the one or more vehicle objects and one or more human objects corresponding to a portion of the image, the single AI model having a substantially same number of nodes and layers and being substantially a same size in memory as an AI model that detects only the one or more vehicle objects or detects only the one or more human objects, the single detection function and the single AI model remaining resident in memory, between processing the image and processing a second image, wherein when the second image is received, the second image is processed without reloading the single AI model into the memory;
for each of the one or more vehicle objects, processing a corresponding portion of the image to determine a plurality of properties of a vehicle object based on a first weight of the single AI model, and to generate annotations of the one or more vehicle objects in the corresponding portion of the image with the plurality of properties of the vehicle object;
for each of the one or more human objects, processing a corresponding portion of the image to determine a plurality of properties of a human object based on a second weight of the single AI model, and to generate annotations of the one or more human objects in the corresponding portion of the image with the plurality of properties of the human object;
assembling annotations of the one or more vehicle objects and annotations of one or more human objects as metadata of the image, wherein the metadata of the image includes metadata that describes the portion of the image within the image that corresponds to the one or more vehicle objects and the properties of the one or more vehicle objects and metadata that describes the portion of the image within the image that corresponds to the one or more human objects and the properties of the one or more human objects, the properties of the one or more vehicle objects including a type of vehicle, a license number of the vehicle, and a color of the vehicle, the properties of the one or more human objects including an approximate age of a person, a skin tone, an eye color, a hair color, a gender, and facial landmarks of the person's face; and
transmitting the image, with annotations of the one or more vehicle objects and the annotations of one or more human objects being assembled, to a service or application that utilizes an annotated image to perform a function of the service or application.
|