US 12,243,241 B2
Tracking of multiple objects using neural networks, local memories, and a shared memory
Cosmin Ionut Bercea, Haar (DE)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Filed by Robert Bosch GmbH, Stuttgart (DE)
Filed on Mar. 16, 2022, as Appl. No. 17/696,050.
Claims priority of application No. 10 2021 202 934 .5 (DE), filed on Mar. 25, 2021.
Prior Publication US 2022/0309681 A1, Sep. 29, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. G06T 7/00 (2017.01); G06F 9/54 (2006.01); G06N 3/045 (2023.01); G06T 7/20 (2017.01)
CPC G06T 7/20 (2013.01) [G06F 9/544 (2013.01); G06N 3/045 (2023.01); G06T 2207/20084 (2013.01); G06T 2207/30236 (2013.01); G06T 2207/30252 (2013.01)] 13 Claims
OG exemplary drawing
 
1. A method for tracking and/or characterizing multiple objects in a sequence of images, comprising the following steps:
assigning a respective neural network to each object of the multiple objects to be tracked;
providing a memory that is shared by all of the respective neural networks;
providing a respective local memory for each respective neural network;
supplying images from the sequence, and/or details of these images, to each of the respective neural networks;
during processing of each image and/or image detail by a neural network of the respective neural networks, generating an address vector from at least one processing product of the neural network;
based on the address vector, writing at least one further processing product of the neural network into the shared memory and/or into the local memory, and/or reading out data from the shared memory and/or local memory and further processing the read out data by the neural network; and
delivering, as and output, by each respective neural network, positions of the object to which the respective neural network is assigned in the images or image details supplied to the respective neural network, and/or information concerning behavior or other sought properties of the object to which the respective neural network is assigned;
wherein the shared memory and/or at least one local memory of the neural network is configured to map an address vector of address components, via differentiable operations, onto one or multiple memory locations, and to read data from the memory locations or write data into the memory locations.