Method and system for tire tread pattern image segmentation

Patent metadata

Field	Value
Publication number	US-12530883-B2
Application number	US-202318328837-A
Country	US
Kind code	B2
Filing date	Jun 5, 2023
Priority date	Jun 5, 2023
Publication date	Jan 20, 2026
Grant date	Jan 20, 2026

Abstract

The invention provides a computer-implemented method and a system for automatically segmenting thermal images of a tire's footprint in order to identify components of the tire's tread pattern. The invention relies on a convolutional neural network model and on digital image pre-filtering. It is capable of accurately segmenting thermal footprint images of tires in either straight rolling or cornering conditions, which allows for the accurate extraction of temperature data for each identified tread pattern component.

First claim

The invention claimed is:

1 . A computer-implemented method for identifying components of a tire's tread pattern in an input thermal footprint image of a tire acquired from underneath a surface that is transparent to infrared radiation, while the tire is rolling thereon, the method comprising the steps of
providing a convolutional neural network model in a memory element, which is trained to identify the components of a tire's tread pattern in an input thermal footprint image of a tire; executing the convolutional neural network model to identify components of a tire's tread pattern in the input thermal footprint image of a tire, thereby generating a segmentation mask of the input thermal footprint image;
wherein executing the convolutional neural network model comprises the steps of:
associating a resolution value with each of a plurality of consecutive processing stages forming the convolutional neural network model, each processing stage comprising an encoder block and an associated decoder block;
processing, at an encoder block of at least one processing stage, an encoder input having the resolution value associated with the processing stage, into an output having the same resolution value, which is forwarded to an associated decoder block, and which is down-sampled to a lower resolution value for processing by a processing stage associated with the lower resolution value;
processing, at a decoder block of said at least one processing stage, a decoder input comprising a concatenation of the output forwarded by an associated encoder block and of an up-sampled output generated by a processing stage associated with a lower resolution value, into a processing stage output having the resolution value associated with the decoder block's processing stage;

and wherein processing at the encoder block and at the decoder block comprises
processing the respective input into a plurality of filtered intermediate outputs by using an atrous convolution step, wherein a filtering kernel is used at a corresponding plurality of different kernel dilution rates, and
computing the respective output by generating an average of the filtered intermediate outputs.

2 . The computer-implemented method according to claim 1 , wherein the step of providing a convolutional neural network model comprises the preliminary steps of:
providing a plurality of thermal footprint training images of at least one tire, in which known components of the respective tire's tread pattern are identified; executing the convolutional neural network model using each of the thermal footprint training images as an input thermal footprint image, to identify components of a tire's tread pattern in the thermal footprint training images, thereby generating a segmentation mask of the corresponding thermal footprint training images, and iteratively updating trainable parameters of the convolutional neural network model so as to minimize an error between the components of a tire's tread pattern that are identified by the convolutional neural network model, and the known components of the tire's tread pattern, which are identified in the corresponding thermal footprint training image of the tire;
thereby training the convolutional neural network model.

3 . The computer-implemented method according to claim 1 , wherein the input thermal footprint image of a tire is pre-processed using an edge enhancing digital image filter prior to executing the convolutional neural network model to identify components of a tire's tread pattern in the pre-processed input thermal footprint image.

4 . The computer-implemented method according to claim 1 , wherein the input thermal footprint image of a tire is pre-processed using a filter bank comprising a plurality of Gabor filters, wherein each Gabor filter is characterized by a different combination of orientation and scale.

5 . The computer-implemented method according to claim 1 , wherein the step of executing the convolutional neural network model comprises pre-filtering the encoder input of the first processing stage at a pre-filtering stage that includes an edge enhancing digital image filter and a digital image filter having trainable parameters for pre-filtering the input thermal footprint image of a tire, and wherein the trainable parameters of the digital image filter are determined during a training step.

6 . The computer-implemented method according to claim 5 , wherein the edge enhancing digital image filter comprises a filter bank of Gabor filters, wherein each Gabor filter is characterized by a different combination of orientation and scale, and wherein the digital image filter having trainable parameters comprises a filter bank of digital image filters having trainable parameters.

7 . The computer-implemented method according to claim 1 , wherein processing the respective input at the encoder block and at the decoder block of a processing stage of the convolutional neural network model into a plurality of filtered intermediate outputs by using an atrous convolution step comprises processing the respective input into three filtered intermediate outputs using an atrous convolution step, wherein a same filtering kernel of size 7 times 7 pixels is used at the three kernel dilution rates of 1, 2 and 4.

8 . The computer-implemented method according to claim 1 , wherein the convolutional neural network model comprises three processing stages.

9 . The computer-implemented method according to claim 1 , wherein the generated segmentation mask is a multi-class segmentation mask comprising at least three different classes.

10 . The computer-implemented method according to claim 1 , wherein the step of executing the convolutional neural network model comprises processing the output of the decoder block of the first processing stage, using a final activation function of the SoftMax type, for classifying the output of the decoder block of the first processing stage into a plurality of at least three different classes, thereby identifying a corresponding plurality of component types of a tire's tread pattern.

11 . The computer-implemented method according to claim 1 , wherein the trained convolutional neural network model is capable of identifying at least background, ribs and groves of a tire's tread pattern in an input thermal footprint image of the corresponding tire.

12 . The computer-implemented method according to claim 1 , wherein the input thermal footprint image of a tire is acquired using an infrared imaging device.

13 . The computer-implemented method according to claim 1 , wherein the input thermal footprint image of a tire is pre-segmented into two classes, wherein a first class comprises the tire's footprint, and wherein a second class comprises background content.

14 . The computer-implemented method according to claim 1 , further comprising a step of extracting temperature information of a component of a tire's tread pattern from an input thermal footprint image of the tire, based on the generated segmentation mask, which identifies the location of said component in the input thermal footprint image of the tire.

15 . A computer-implemented method for identifying components of a tire's tread pattern in an input thermal footprint video of a tire comprising a sequence of input thermal footprint images of a tire acquired from underneath a surface that is transparent to infrared radiation while the tire is rolling thereon, the method comprising the steps of
providing a convolutional neural network model in a memory element, which is trained to identify the components of a tire's tread

Assignees

• Goodyear Tire & Rubber
• Luxembourg Institute Of Science And Tech

Inventors

• Fehr Duc Alexandre
• Petry Frank
• Nava Velazco Uriel Rodrigo

Classifications

• G06T5/60

  using machine learning, e.g. neural networks · CPC title

• G06T2207/20192

  Edge enhancement; Edge preservation · CPC title

• G06T2207/20024

  Filtering details · CPC title

• G06T2207/10048

  Infrared image · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title