Granted Patent B2

US 12,536,815 · App. 17/939,063 · Granted Jan 27, 2026

Inventors: Georgiana Negishi (Tokyo, JP); Szabolcs Fulop (Brasov, RO); Tudor Topoleanu (Brasov, RO); Paul George (Troy, MI)

Assignee: Tobii Technologies Limited

G06V20/597G06V10/25G06V10/82G06V40/10

Abstract

Claims (23)

1 . A vehicle cabin monitoring system comprising:

a processor configured to obtain a sequence of images from an image acquisition device having a field of view including at least a portion of a vehicle cabin, said processor comprising one or more image processing components configured to obtain image information from an image and to make an inference based on said image information, the inferential ability of at least one of said one or more image processing components being governed by core processing parameters;

a vehicle geometry store, the store encoding a spatial relationship between said image acquisition device and at least one user interactive control for said vehicle, so that a location of the or each user interactive control within the field of view of the image acquisition device can be determined;

a customization profile for storing:

one or more annotated images, each annotated image being associated with at least one annotation indicating a ground truth for an associated region of said annotated image; and

a plurality of core processing parameters for at least one image processing component of said processor;

wherein the customization profile is configured to store at least one customized threshold for a user in relation to an image processing component for testing a classification provided by said image processing component and modify said customized threshold before re-training said image processing component and determining a new ground truth for said image processing component, wherein the re-training is initiated in response to a count of accumulated ground truths reaching the customized threshold, wherein the customized threshold is a predefined number of ground truths that need to be accumulated for each classification;

the system being:

responsive to user interaction with a user interactive control of said vehicle within said field of view of said image acquisition device for storing an image acquired by said image acquisition device at or around the time of said interaction in said customization profile with an annotation indicating

a ground truth for an associated region of said image according to said interaction; and

configured to use one or more annotated images from said customization profile for re-training an image processing component of said processor and for storing updated core processing parameters produced by said re-training in said customization profile for use by said re-trained image processing component in processing subsequently acquired images.

2 . The system of claim 1 wherein the user interactive controls include: controls for adjusting mirrors, controls for vehicle wipers, a center console, a horn, a steering wheel, lamp controls, seat controls, a gear shift, an indicator, a brake, a door handle or a seat buckle.

3 . The system of claim 1 wherein the annotations comprise any of: eye gaze angle, a hand indication, a body indication, a body pose indication or an emotion indication for a region of interest of an image.

4 . The system of claim 1 configured to communicate with one or more non-interactive vehicle components in order to determine vehicle state.

5 . The system of claim 1 wherein said one or more non-interactive vehicle components include a tachometer.

6 . The system of claim 1 wherein said image processing component comprises a classifier with at least one fully connected classification layer and wherein re-training is limited to adjusting weights for a subset of said fully connected classification layers.

7 . The system of claim 1 wherein said subset comprises a final set of fully connected classification layers.

8 . The system of claim 1 wherein said system is configured to perform said re-training within said vehicle.

9 . The system of claim 1 wherein said system is configured to provide said one or more annotated images for an image processing component to an external re-training system and to receive updated core processing parameters for said image processing component from said external re-training system.

10 . The system of claim 1 comprising a customization profile for each driver of the vehicle, the system being responsive to being provided with an identity of the driver for storing one or more annotated images and a plurality of core processing parameters for at least one image processing component of said processor in the customization profile for the identified driver.

11 . The system of claim 1 wherein the system is responsive said user interaction lwith a user interactive control of said vehicle within said field of view of said image acquisition device and a corresponding image processing component failing to detect said interaction for storing said image acquired by said image acquisition device at or around the time of said interaction in said customization profile with an annotation indicating a ground truth for an associated region of said image according to said interaction.

12 . The system of claim 1 being responsive to receiving a plurality of core processing parameters for at least one image processing component of said processor for deleting one or more annotated images used for generating said core processing parameters from said customization profile.

13 . The system of claim 1 wherein modifying said customized threshold comprises reducing a value of said customized threshold in response to determining a previous ground truth is incorrect.

Assignments (2)

Continuity (1)

References Cited (23)