	US 12,444,204 B2
	Method and system for predicting the presence of environmental conditions in image data
Ognjan Hedberg, Stenungsund (SE)
Assigned to ZENUITY AB, Gothenburg (SE)
Appl. No. 17/914,018
Filed by ZENUITY AB, Gothenburg (SE)
PCT Filed Mar. 31, 2020, PCT No. PCT/EP2020/059090 § 371(c)(1), (2) Date Sep. 23, 2022, PCT Pub. No. WO2021/197584, PCT Pub. Date Oct. 7, 2021.
Prior Publication US 2023/0138969 A1, May 4, 2023
Int. Cl. G06V 20/56 (2022.01); B60S 1/08 (2006.01); G06V 10/94 (2022.01); G06V 20/70 (2022.01)

CPC G06V 20/56 (2022.01) [B60S 1/0844 (2013.01); G06V 10/95 (2022.01); G06V 20/70 (2022.01)]

19 Claims

1. A method for automated supervised training of a learning algorithm for detecting an environmental condition in a traffic environment, the method comprising:

obtaining image data over a time period originating from a camera device arranged on at least one vehicle;

wherein the obtained image data comprises a plurality of labelled images, wherein the plurality of labelled images having labels indicative of:

an activation signal outputted by an environmental sensor of the at least one vehicle, the activation signal being indicative of a presence of an environmental condition and arranged to control an operation of a control system of the at least one vehicle,

a deactivation signal outputted by the environmental sensor of the at least one vehicle, the deactivation signal being indicative of an ending of the environmental condition and arranged to control the operation of the control system of the at least one vehicle, and

a user-input signal outputted by a user-input device of the at least one vehicle, the user input-signal being indicative of an override of the activation signal or the deactivation signal of the environmental sensor of the at least one vehicle and arranged to control the operation of the control system of the at least one vehicle; and

training a learning algorithm based on the obtained plurality of labelled images, in order to form a trained learning algorithm configured to predict the presence of the environmental condition in the obtained image data.

8. A system for automated supervised training of a learning algorithm for detecting an environmental condition in a traffic environment, the system comprising:

a control circuitry configured to:

obtain image data over a time period originating from a camera device arranged on at least one vehicle;

wherein the obtained image data comprises a plurality of labelled images, wherein the plurality of labelled images having labels indicative of:

train a learning algorithm based on the obtained plurality of labelled images, in order to form a trained learning algorithm configured to predict the presence of the environmental condition in the obtained image data.

13. A method for in-vehicle generation of training data for automated supervised training of a learning algorithm for detecting an environmental condition in a traffic environment, the method comprising:

obtaining image data over a time period from a camera device arranged on a vehicle, the obtained image data comprising a plurality of images of a surrounding environment of the vehicle over the time period;

obtaining a first signal indicative of an activation signal outputted by an environmental sensor of the vehicle during the time period, the activation signal being indicative of a presence of an environmental condition and arranged to control an operation of a control system of the vehicle;

obtaining a second signal indicative of a deactivation signal outputted by the environmental sensor of the vehicle during the time period, the deactivation signal being indicative of an ending of the environmental condition and arranged to control the operation of the control system of the vehicle;

obtaining a third signal indicative of a user-input signal outputted by a user-input device of the vehicle during the time period, the user input-signal being indicative of an override of the activation signal or the deactivation signal of the environmental sensor of the at least one vehicle and arranged to control the operation of the control system of the at least one vehicle; and

transmitting a set of the plurality of images to a remote entity, the set of images being associated with timings of the activation signal, deactivation signal, and the user-input signal over the time period.

17. A system for in-vehicle generation of training data for automated supervised training of a learning algorithm for detecting an environmental condition in a traffic environment, the system comprising:

a control circuitry configured to:

obtain image data over a time period from a camera device arranged on a vehicle, the obtained image data comprising a plurality of images of a surrounding environment of the vehicle over the time period;

obtain a first signal indicative of an activation signal outputted by an environmental sensor of the vehicle during the time period, the activation signal being indicative of a presence of an environmental condition and arranged to control an operation of a control system of the vehicle;

obtain a second signal indicative of a deactivation signal outputted by the environmental sensor of the vehicle during the time period, the deactivation signal being indicative of an ending of the environmental condition and arranged to control the operation of the control system of the vehicle;

obtain a third signal indicative of a user-input signal outputted by a user-input device of the vehicle during the time period, the user input-signal being indicative of an override of the activation signal or the deactivation signal of the environmental sensor of the at least one vehicle and arranged to control the operation of the control system of the at least one vehicle; and

transmit a set of the plurality of images to a remote entity, the set of images being associated with timings of the activation signal, deactivation signal, and the user-input signal over the time period.