Method of automatic tire inspection and system thereof

1. A computerized method of automatic tire inspection, the method being performed by a computer and comprising:

obtaining at least one image acquired by an image acquisition device, the at least one image capturing a wheel of a vehicle;

segmenting the at least one image into one or more image segments corresponding to one or more mechanical components of the wheel, wherein the one or more image segments include a tire image segment corresponding to a tire of the wheel, the tire image segment characterized by a curved shape;

straightening the tire image segment from the curved shape to a straight shape, giving rise to a straight tire segment;

identifying text marked on the tire from the straight tire segment, comprising:

detecting locations of a plurality of text portions on the straight tire segment using a text detection module, and

recognizing text content for each of the text portions using a text recognition module operatively connected to the text detection module; and

analyzing the recognized text content based on one or more predefined rules indicative of association between text content of different text portions at given relative locations, giving rise to a text analysis result indicative of condition of the tire.

2. The computerized method according to claim 1 , wherein the at least one image is segmented using a segmentation deep learning model, the segmentation deep learning model being trained using a training dataset comprising a set of training wheel images each segmented and labeled according to one or more mechanical components comprised therein.

3. The computerized method according to claim 1 , wherein the one or more image segments include a rim image segment corresponding to a rim of the wheel.

4. The computerized method according to claim 1 , wherein the text detection module comprises a deep learning neural network trained using a training dataset comprising a set of straight tire segments each labeled with one or more bounding boxes containing one or more respective text portions.

5. The computerized method according to claim 1 , wherein the text recognition module comprises a deep learning neural network trained using a training dataset comprising a set of image portions each containing a respective text portion and labeled with text content comprised in the respective text portion.

6. The computerized method according to claim 1 , further comprising analyzing the tire image segment or the straight tire segment and computing a ratio between upper sidewall height and lower sidewall height of the tire, wherein the ratio is indicative of pressure condition of the tire.

7. The computerized method according to claim 1 , further comprising performing anomaly detection on at least the tire image segment for identifying one or more anomalies indicative of potential damages on the tire.

8. The computerized method according to claim 7 , wherein the anomaly detection is performed using a unsupervised deep learning model, the unsupervised deep learning model being trained using a training dataset comprising a set of tire image segments without anomaly so that the trained model is capable of extracting features representative of a tire image segment without anomaly.

9. The computerized method according to claim 8 , wherein the trained deep learning model is used in runtime to detect anomalies by extracting features representative of a runtime tire image segment, and determining whether the extracted features are consistent with the features representative of a tire image segment without anomaly.

10. The computerized method according to claim 7 , wherein the anomaly detection is performed using a supervised deep learning model, the supervised deep learning model being trained using a training dataset comprising a set of tire image segments including one or more tire image segments with anomaly and one or more tire image segments without anomaly so that the trained supervised deep learning model is capable of identifying, in runtime, one or more runtime tire image segments with anomaly.

11. The computerized method according to claim 1 , wherein the condition of the tire includes improper tire usage and/or improper tire installation.

12. A computerized system for automatic tire inspection, the system comprising a processor and memory circuitry (PMC) configured to:

obtain at least one image acquired by an image acquisition device, the at least one image capturing a wheel of a vehicle;

segment the at least one image into one or more image segments corresponding to one or more mechanical components of the wheel, wherein the one or more image segments include a tire image segment corresponding to a tire of the wheel, the tire image segment characterized by a curved shape;

straighten the tire image segment from the curved shape to a straight shape, giving rise to a straight tire segment;

identify text marked on the tire from the straight tire segment, the identifying comprising:

detecting locations of a plurality of text portions on the straight tire segment using a text detection module, and

recognizing text content for each of the text portions using a text recognition module operatively connected to the text detection module; and

analyze the recognized text content based on one or more predefined rules indicative of association between text content of different text portions at given relative locations, giving rise to a text analysis result indicative of condition of the tire.

13. The computerized system according to claim 12 , wherein the at least one image is segmented using a segmentation deep learning model, the segmentation deep learning model being trained using a training dataset comprising a set of training wheel images each segmented and labeled according to one or more mechanical components comprised therein.

14. The computerized system according to claim 12 , wherein the text detection module comprises a deep learning neural network trained using a training dataset comprising a set of straight tire segments each labeled with one or more bounding boxes containing one or more respective text portions.

15. The computerized system according to claim 12 , wherein the text recognition module comprises a deep learning neural network trained using a training dataset comprising a set of image portions each containing a respective text portion and labeled with text content comprised in the respective text portion.

16. The computerized system according to claim 12 , wherein the PMC is further configured to perform anomaly detection on at least the tire image segment for identifying one or more anomalies indicative of potential damages on the tire.

17. The computerized system according to claim 16 , wherein the PMC is configured to perform anomaly detection using a unsupervised deep learning model, the unsupervised deep learning model being trained using a training dataset comprising a set of tire image segments without anomaly so that the trained model is capable of extracting features representative of a tire image segment without anomaly.

18. The computerized system according to claim 17 , wherein the trained deep learning model is used in runtime to detect anomalies by extracting features representative of a runtime tire image segment, and determining whether the extracted features are consistent with the features representative of a tire image segment without anomaly.

19. The computerized system according to claim 16 , wherein the PMC is configured to perform anomaly detection using a supervised deep learning model, the supervised deep learning model being trained using a training dataset comprising a set of tire image segments including one or more tire image segments with anomaly and one or more tire image segments without anomaly so that the trained supervised deep learning model is capable of identifying, in runtime, one or more runtime tire image segments with anomaly.

20. A non-transitory computer readable storage medium tangibly embodying a program of instructions that, when executed by a computer, causing the computer to perform a method of automatic tire inspection, the method comprising:

obtaining at least one image acquired by an image acquisition device, the at least one image capturing a wheel of a vehicle;

segmenting the at least one image into one or more image segments corresponding to one or more mechanical components of the wheel, wherein the one or more image segments include a tire image segment corresponding to a tire of the wheel, the tire image segment characterized by a curved shape;

straightening the tire image segment from the curved shape to a straight shape, giving rise to a straight tire segment;

identifying text marked on the tire from the straight tire segment, comprising:

detecting locations of a plurality of text portions on the straight tire segment using a text detection module, and

recognizing text content for each of the text portions using a text recognition module operatively connected to the text detection module; and

analyzing the recognized text content based on one or more predefined rules indicative of association between text content of different text portions at given relative locations, giving rise to a text analysis result indicative of condition of the tire.