US 12,311,954 B2
Method and system for robust identification of a vehicle occupant
Mohamed Kari, Essen (DE); David Bethge, Stuttgart-Feuerbach (DE); and Satiyabooshan Murugaboopathy, Dresden (DE)
Assigned to Dr. Ing. h.c. F. Porsche Aktiengesellschaft, (DE)
Filed by Dr. Ing. h.c. F. Porsche Aktiengesellschaft, Stuttgart (DE)
Filed on Oct. 17, 2022, as Appl. No. 17/967,019.
Claims priority of application No. 10 2021 129 103.8 (DE), filed on Nov. 9, 2021.
Prior Publication US 2023/0147024 A1, May 11, 2023
Int. Cl. B60W 40/08 (2012.01); G06N 3/045 (2023.01)
CPC B60W 40/08 (2013.01) [G06N 3/045 (2023.01); B60W 2420/403 (2013.01); B60W 2420/408 (2024.01); B60W 2540/043 (2020.02)] 11 Claims
OG exemplary drawing
 
1. A method for robust identification of occupants in a vehicle, wherein the vehicle has a first interior sensor and a second interior sensor that are each disposed in a vehicle interior, said method comprising:
carrying out a Deep Canonical Correlation Analysis to identify a respective vehicle occupant among a group of vehicle users with a first neural network associated with the first interior sensor and a second neural network associated with the second interior sensor, wherein the first neural network and the second neural network have a symmetrical structure comprising at least three hidden layers and are each determined by a plurality of network parameters, wherein a respective data vector from a respective measuring operation of the first interior sensor and the second interior sensor is formed as a respective input for the respective neural network, wherein a respective representation is issued by the respective neural network as a respective output, wherein a loss function is formed via a correlation between the respective representations, wherein the loss function is minimal when the correlation is maximum,
(a) wherein, in a first step serving as an initialization, each vehicle occupant from the group of vehicle users is authenticated against the output of the respective neural networks by:
(i) providing a respective pair of data vectors as a respective input for each vehicle occupant in a predetermined number of training runs by way of a respective simultaneous measuring operation by the first interior sensor and the second interior sensor,
(ii) optimizing the plurality of network parameters of the respective neural networks with regard to a minimum value of the loss function by way of a gradient-based process, and,
(iii) after completion of the training runs, recalculating the representations of the output for all data vector pairs with the optimized network parameters, forming a respective averaged representation and storing the respective averaged representation in a user account associated with the respective vehicle occupant, and
(b) wherein, in a second step serving as a productive use, the respective vehicle occupant is identified by:
(i) providing at least one data vector of the data vector pair as an input to the respective associated trained neural network via a respective measuring operation carried out each time the vehicle occupant reenters the interior of the vehicle, and
(ii) identifying the respective vehicle occupant by comparing the at least one output with the respective averaged representations stored in the user accounts.