Tooth-position recognition system
A tooth-position recognition system includes an electronic device and a calculation device. The electronic device includes a first camera. The first camera is configured to capture a plurality of tooth images. The calculation device includes a second camera and a processor. The second camera is configured to capture a user image. The processor is configured to receive the tooth images, compare the corresponding position of each pixel in each tooth image to generate a depth map, and input the tooth images, the depth map, and a plurality of first tooth-region identifiers into a tooth deep-learning model. The tooth deep-learning model outputs a plurality of deep-learning probability values that are the same in number as the first tooth-region identifiers. The processor inputs the user image and the plurality of second tooth-region identifiers into a user-image deep-learning model.
1. A tooth-position recognition system, comprising:
an electronic device, comprising:
a first camera, configured to capture a plurality of tooth images; and
a calculation device, comprising:
a second camera, configured to capture a user image; and
a processor, configured to receive the tooth images, compare a corresponding position of each pixel in each tooth image to generate a depth map, input the tooth images, the depth map, and a plurality of first tooth-region identifiers into a tooth deep-learning model; wherein the tooth deep-learning model outputs a plurality of deep-learning probability values that are the same in number as the first tooth-region identifiers; wherein the processor inputs the user image and the plurality of second tooth-region identifiers into a user-image deep-learning model, the user-image deep-learning model outputs a left region probability value and a right region probability value; wherein the processor treats the deep-learning probability values, the left regional probability value, and the right regional probability value as a plurality of feature values, and inputs the feature values and a plurality of third tooth-region identifiers into a multi-layer perceptron classifier, and the multi-layer perceptron classifier outputs a tooth-position probability corresponding to the tooth images.
2. The tooth-position recognition system of claim 1 , further comprising:
an inertial measurement unit (IMU), configured to measure posture information and motion track of the electronic device;
wherein the processor inputs the posture information, the motion track and a plurality of fourth tooth-region identifiers into an IMU-movement deep-learning model, and the IMU-movement deep-learning model outputs a plurality of quadrant probability values.
3. The tooth-position recognition system of claim 2 , wherein the processor treats the deep-learning probability values, the left regional probability value, the right regional probability value, and the quadrant probability values as feature values, and inputs the feature values and the third tooth-region identifiers into a multi-layer perceptron classifier, and the multi-layer perceptron classifier outputs a tooth-position probability corresponding to the tooth images.
4. The tooth-position recognition system of claim 1 , wherein the tooth images include an R channel array image, a G channel array image, and a B channel array image.
5. The tooth-position recognition system of claim 2 , wherein the tooth deep-learning model and the user-image deep-learning model are each implemented by a convolutional neural network (CNN) model, and the IMU motion deep learning model is implemented by a recurrent neural network (RNN) model.
6. A tooth-position recognition system, comprising:
an electronic device, comprising:
a first camera, configured to capture a plurality of tooth images; and
an inertial measurement unit (IMU), configured to measure the posture information and the motion track of the electronic device; and
a calculation device, comprising:
a processor, configured to receive the tooth images, compare the corresponding position of each pixel in each tooth image to generate a depth map, input the tooth images, the depth map, and a plurality of first tooth-region identifiers into a tooth deep-learning model; wherein the tooth deep-learning model outputs a plurality of deep-learning probability values that are the same in number as the first tooth-region identifiers; wherein the processor inputs the user image and the plurality of second tooth-region identifiers into an IMU-movement deep-learning model, and the IMU-movement deep-learning model outputs a plurality of quadrant probability values; wherein the processor treats the deep-learning probability values and the quadrant probability values as feature values, and inputs the feature values and a plurality third tooth-region identifiers into the multi-layer perceptron classifier, and the multi-layer perceptron classifier outputs a tooth-position probability corresponding to the tooth images.
7. The tooth-position recognition system of claim 6 , which further comprises:
a second camera, configured to capture a user image;
wherein the processor inputs the user image and a plurality of fourth tooth-region identifiers into a user-image deep-learning model, the user-image deep-learning model outputs a left region probability value and a right region probability value.
8. The tooth-position recognition system of claim 7 , wherein the processor treats the deep-learning probability values, the left regional probability value and the right regional probability value as feature values, and inputs the feature values and the third tooth-region identifiers into the multi-layer perceptron classifier, and the multi-layer perceptron classifier outputs a tooth-position probability corresponding to the tooth images.
9. The tooth-position recognition system of claim 8 , wherein the tooth images include an R channel array image, a G channel array image, and a B channel array image.
10. The tooth-position recognition system of claim 7 , wherein the tooth deep-learning model and the user-image deep-learning model are each implemented by a convolutional neural network (CNN) model, and the IMU motion deep learning model is implemented by a recurrent neural network (RNN) model.