Detection circuit and associated detection method

1 . A detection circuit, comprising:

a neural network module circuit, configured to receive an image to generate an output tensor, wherein the output tensor comprises position information of a specific object and distance adjustment information, wherein the distance adjustment information is a dimensionless scaling factor; and

a calculation circuit, coupled to the neural network module, configured to calculate an initial distance between an image capture device and the specific object according to the position information of the specific object, and generate an estimated distance according to the initial distance and the distance adjustment information;

wherein the specific object is a person, the output tensor comprises the position information of the person and the distance adjustment information; and the calculation circuit calculates the initial distance between the image capture device and the specific object according to the position information, a default width and a default height of the person;

wherein the calculation circuit multiplies the distance adjustment information by the initial distance to generate the estimated distance.

2 . The detection circuit of claim 1 , wherein during a training phase of the neural network module, the neural network module circuit receives a plurality of training images; and for each of the plurality of training images, the neural network module circuit receives the training image to generate a training output tensor, and a loss function operation is performed on the training output tensor and a training parameter for calibrating parameters in the neural network module; wherein the plurality of training images respectively comprise persons with different actions/postures.

3 . The detection circuit of claim 2 , wherein the training parameter comprises another distance adjustment information calculated according to a real distance between the image capture device and the person and real position information of the person, wherein the another distance adjustment information serves as a training target in the training phase of the neural network module.

4 . A detection method, comprising:

using a neural network module circuit to receive an image to generate an output tensor, wherein the output tensor comprises position information of a specific object and distance adjustment information, wherein the distance adjustment information is a dimensionless scaling factor; and

calculating an initial distance between an image capture device and the specific object according to the position information of the specific object; and

generating an estimated distance according to the initial distance and the distance adjustment information;

wherein the specific object is a person, the output tensor comprises the position information of the person and the distance adjustment information; and the step of calculating the initial distance between the image capture device and the specific object according to the position information of the specific object comprises:

calculating the initial distance between the image capture device and the specific object according to the position information, a default width and a default height of the person;

wherein the step of generating the estimated distance according to the initial distance and the distance adjustment information comprises:

multiplying the distance adjustment information by the initial distance to generate the estimated distance.

5 . The detection method of claim 4 , further comprising:

during a training phase of the neural network module,

receiving a plurality of training images, wherein the plurality of training images respectively comprise persons with different actions/postures;

for each of the plurality of training images, using the neural network module circuit to receive the training image to generate a training output tensor; and

performing a loss function operation on the training output tensor and a training parameter, for calibrating parameters in the neural network module.

6 . The detection method of claim 5 , wherein the training parameter comprises another distance adjustment information calculated according to a real distance between the image capture device and the person and real position information of the person, wherein the another distance adjustment information serves as a training target in the training phase of the neural network module.

7 . The detection circuit of claim 3 , wherein the another distance adjustment information is a ratio of the real distance to an initial training distance calculated according to the real position information.

8 . The detection method of claim 6 , wherein the another distance adjustment information is a ratio of the real distance to an initial training distance calculated according to the real position information.