Controller, method, and computer program for controlling vehicle

1. A vehicle controller comprising:

a processor configured to:

detect an object region from an image obtained by a camera mounted on a vehicle, the camera being configured to capture surroundings of the vehicle, the object region representing a target vehicle traveling in an area around the vehicle;

input characteristics extracted from the object region in the image into a light-state classifier to identify a light state that is the state of a signal light of the target vehicle, the light-state classifier having been trained to identify the light state of the target vehicle;

determine a first control value, based on the result of identification of the light state of the target vehicle and the positional relationship between the vehicle and the target vehicle, in accordance with a predetermined rule for controlling the vehicle to avoid a collision between the vehicle and the target vehicle;

input the result of identification of the light state of the target vehicle into a control command classifier to determine a second control value for controlling the vehicle to avoid a collision between the vehicle and the target vehicle, the control command classifier having been trained to output the second control value;

determine an integrated control value for avoiding a collision between the vehicle and the target vehicle, based on the difference between the first and second control values or the absolute value of the second control value; and

control travel of the vehicle according to the integrated control value.

2. The vehicle controller according to claim 1 , wherein the processor outputs the first and second control values, respectively, every predetermined time interval, and

in the case that the absolute value of the difference between the first and second control values has not been greater than a certain value over a period not shorter than a predetermined time threshold, the processor determines an average of any one of first control values and any one of second control values outputted in the period as the integrated control value.

3. The vehicle controller according to claim 1 , wherein the processor outputs the first and second control values, respectively, every predetermined time interval, and

in the case that the absolute value of the difference between the first and second control values has not been less than a certain value over a period not shorter than a predetermined time threshold, the processor determines the greater of the absolute value of an average of first control values outputted in the period and the absolute value of an average of second control values outputted in the period as the integrated control value.

4. The vehicle controller according to claim 1 , wherein when the absolute value of the second control value is not less than a predetermined upper limit, the processor determines the predetermined upper limit or the first control value as the integrated control value.

5. The vehicle controller according to claim 1 , wherein the processor determines the second control value by inputting, into the control command classifier, the result of identification of the light state of the target vehicle and at least one of the following: the positional relationship between the vehicle and the target vehicle, a relative speed between the vehicle and the target vehicle, relative acceleration/deceleration between the vehicle and the target vehicle, the size of the object region, and the position of the object region.

6. The vehicle controller according to claim 1 , wherein the light-state classifier includes an input layer into which characteristics extracted from the object region are inputted, an output layer that outputs the light state of the target vehicle, and at least one intermediate layer between the input layer and the output layer, the at least one intermediate layer being provided sequentially from the input layer toward the output layer and each executing a predetermined operation on output from a layer closer to the input layer than the intermediate layer, and

the processor determines the second control value by inputting the result of identification of the light state of the target vehicle and the result of the predetermined operation by any one of the at least one intermediate layer of the light-state classifier into the control command classifier.

7. The vehicle controller according to claim 1 , wherein the light-state classifier has a recursive structure, and

the processor identifies the light state of the target vehicle by inputting the characteristics extracted from the object region in the image and characteristics extracted from the object region in each of one or more past images obtained by the camera earlier than the image into the light-state classifier in chronological order.

8. The vehicle controller according to claim 1 , wherein the light-state classifier is a convolutional neural network including a convolution layer that executes a convolution operation in a temporal direction, and

the processor identifies the light state of the target vehicle by inputting the characteristics extracted from the object region in the image and characteristics extracted from the object region in each of one or more past images obtained by the camera earlier than the image into the light-state classifier.

9. A method for controlling a vehicle, comprising:

detecting an object region representing a target vehicle traveling in an area around the vehicle from an image obtained by a camera mounted on the vehicle, the camera being configured to capture surroundings of the vehicle;

inputting characteristics extracted from the object region in the image into a light-state classifier to identify a light state that is the state of a signal light of the target vehicle, the light-state classifier having been trained to identify the light state of the target vehicle;

determining a first control value, based on the result of identification of the light state of the target vehicle and the positional relationship between the vehicle and the target vehicle, in accordance with a predetermined rule for controlling the vehicle to avoid a collision between the vehicle and the target vehicle;

inputting the result of identification of the light state of the target vehicle into a control command classifier to determine a second control value for controlling the vehicle to avoid a collision between the vehicle and the target vehicle, the control command classifier having been trained to output the second control value;

determining an integrated control value for avoiding a collision between the vehicle and the target vehicle, based on the difference between the first and second control values or the absolute value of the second control value; and

controlling travel of the vehicle according to the integrated control value.

10. A non-transitory recording medium that stores a computer program for controlling a vehicle, the computer program causing a processor mounted on the vehicle to execute a process comprising:

detecting an object region representing a target vehicle traveling in an area around the vehicle from an image obtained by a camera mounted on the vehicle, the camera being configured to capture surroundings of the vehicle;

inputting characteristics extracted from the object region in the image into a light-state classifier to identify a light state that is the state of a signal light of the target vehicle, the light-state classifier having been trained to identify the light state of the target vehicle;

determining a first control value, based on the result of identification of the light state of the target vehicle and the positional relationship between the vehicle and the target vehicle, in accordance with a predetermined rule for controlling the vehicle to avoid a collision between the vehicle and the target vehicle;

inputting the result of identification of the light state of the target vehicle into a control command classifier to determine a second control value for controlling the vehicle to avoid a collision between the vehicle and the target vehicle, the control command classifier having been trained to output the second control value;

determining an integrated control value for avoiding a collision between the vehicle and the target vehicle, based on the difference between the first and second control values or the absolute value of the second control value; and

controlling travel of the vehicle according to the integrated control value.