Method and apparatus of monitoring sensor of driverless vehicle, device and storage medium

1. A computer-implemented method of monitoring a sensor of a driverless vehicle, wherein the computer-implemented method comprises, at the level of a computer device, the steps of:

monitoring a physical state of a monitored sensor, the physical state comprising one or more of position, connection, power-on, signal reception, or data transmission of the monitored sensor;

monitoring a data transmission state of the monitored sensor, the data transmission state comprising one or more of data size, format, frequency, resolution, hash value, timestamp, reflection value, or height value of the monitored sensor;

monitoring output data of the monitored sensor, and using predetermined data to perform cross-validation for the output data;

when (a) a result of monitoring the physical state or the data transmission state indicates a deviation from a predefined range of values for one or more criteria comprised by the physical state or the data transmission state or (b) there is a validation result abnormality, determining that the monitored sensor is getting abnormal, and giving an alarm in a manner comprising at least one of an in-vehicle speech, an image prompt, or a remote notification; and

responsive to determining that the monitored sensor is getting abnormal, recording current traffic environment conditions and key data, and activating a safety response policy of the driverless vehicle;

wherein using the predetermined data to perform the cross-validation for the output data comprises performing cross-validation for the output data transmitted from the monitored sensor by using one of output data of other sensors other than the monitored sensor, a high-precision map, and data previously output by the monitored sensor, or combinations of at least two thereof;

wherein the monitored sensor is a laser radar sensor and performing the cross-validation for output data of the monitored sensor comprises:

acquiring a point cloud recognition result obtained according to output data of the laser radar sensor;

when a vehicle or an obstacle on a road is recognized from the point cloud recognition result but the vehicle or the obstacle is not recognized from both an image recognition result obtained according to output data of an image sensor and an ultrasonic recognition result obtained according to output data of an ultrasonic sensor, determining the laser radar sensor as getting abnormal;

wherein the abnormality of the laser radar sensor is caused by a hacker's attack or interfering in detection by the laser radar sensor with a strong laser, wherein the hacker's attack includes attacking the laser radar sensor by absorbing waves emitted by the laser radar for concealing and disguising.

2. The computer-implemented method according to claim 1 , wherein the monitored sensor is a positioning sensor and performing the cross-validation for the output data of the monitored sensor comprises:

determining a lane where a positioning result output by the positioning sensor lies, according to the high-precision map;

acquiring an image recognition result obtained according to output data of an image sensor, and determining a lane where the driverless vehicle lies according to the image recognition result; and

determining the positioning sensor as getting abnormal when the lane where the positioning result lies is inconsistent with the lane where the driverless vehicle lies;

or,

comparing the positioning result output by the positioning sensor with a positioning result previously output by the positioning sensor;

determining the positioning sensor as getting abnormal when a difference of the two positioning results does not match the travel speed of the driverless vehicle.

3. The computer-implemented method according to claim 1 , wherein the monitored sensor is an image sensor and performing the cross-validation for the output data of the monitored sensor comprises:

acquiring an image recognition result obtained according to output data of the image sensor;

when a traffic light or traffic sign is recognized from the image recognition result but it is determined by querying the high-precision map that the traffic light or traffic sign does not exist at a position where the driverless vehicle lies, determining the image sensor as getting abnormal;

or,

acquiring an image recognition result obtained according to output data of the image sensor;

when a vehicle or pedestrian is recognized from the image recognition result but the vehicle or pedestrian is not recognized from a point cloud recognition result obtained according to output data of a laser radar sensor, determining the image sensor as getting abnormal.

4. A computer device, comprising a memory, a processor and a computer program which is stored on the memory and runs on the processor, wherein the processor, upon executing the computer program, implements a method of monitoring a sensor of a driverless vehicle, wherein the method comprises:

monitoring a physical state of a monitored sensor, the physical state comprising one or more of position, connection, power-on, signal reception, or data transmission of the monitored sensor;

monitoring a data transmission state of the monitored sensor, the data transmission state comprising one or more of data size, format, frequency, resolution, hash value, timestamp, reflection value, or height value of the monitored sensor;

monitoring output data of the monitored sensor, and using predetermined data to perform cross-validation for the output data;

when (a) a result of monitoring the physical state or the data transmission state indicates a deviation from a predefined range of values for one or more criteria comprised by the physical state or the data transmission state or (b) there is a validation result abnormality, determining that the monitored sensor is getting abnormal, and giving an alarm in a manner comprising at least one of an in-vehicle speech, an image prompt, or a remote notification; and

responsive to determining that the monitored sensor is getting abnormal, recording current traffic environment conditions and key data, and activating a safety response policy of the driverless vehicle;

wherein using the predetermined data to perform the cross-validation for the output data comprises performing cross-validation for the output data transmitted from the monitored sensor by using one of output data of other sensors other than the monitored sensor, a high-precision map, and data previously output by the monitored sensor, or combinations of at least two thereof;

wherein when the monitored sensor is a laser radar sensor, performing the cross-validation for output data of the monitored sensor comprises:

acquiring a point cloud recognition result obtained according to output data of the laser radar sensor;

when a vehicle or an obstacle on a road is recognized from the point cloud recognition result but the vehicle or the obstacle is not recognized from both an image recognition result obtained according to output data of an image sensor and an ultrasonic recognition result obtained according to output data of an ultrasonic sensor, determining the laser radar sensor as getting abnormal;

wherein the abnormality of the laser radar sensor is caused by a hacker's attack or interfering in detection by the laser radar sensor with a strong laser, wherein the hacker's attack includes attacking the laser radar sensor by absorbing waves emitted by the laser radar for concealing and disguising.

5. The computer device according to claim 4 , wherein when the monitored sensor is a positioning sensor, performing the cross-validation for the output data of the monitored sensor comprises:

determining a lane where a positioning result output by the positioning sensor lies, according to the high-precision map;

acquiring an image recognition result obtained according to output data of an image sensor, and determining a lane where the driverless vehicle lies according to the image recognition result;

determining the positioning sensor as getting abnormal if the lane where the positioning result lies is inconsistent with the lane where the driverless vehicle lies;

or,

comparing the positioning result output by the positioning sensor with a positioning result previously output by the positioning sensor;

determining the positioning sensor as getting abnormal if a difference of the two positioning results does not match the travel speed of the driverless vehicle.

6. The computer device according to claim 4 , wherein when the monitored sensor is an image sensor, performing the cross-validation for output data of the monitored sensor comprises:

acquiring an image recognition result obtained according to output data of the image sensor;

when a traffic light or traffic sign is recognized from the image recognition result but it is determined by querying the high-precision map that the traffic light or traffic sign does not exist at a position where the driverless vehicle lies, determining the image sensor as getting abnormal;

or,

acquiring an image recognition result obtained according to output data of the image sensor;

when a vehicle or pedestrian is recognized from the image recognition result but the vehicle or pedestrian is not recognized from a point cloud recognition result obtained according to output data of a laser radar sensor, determining the image sensor as getting abnormal.

7. A non-transitory computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, implements a method of monitoring a sensor of a driverless vehicle, wherein the method comprises:

monitoring a physical state of a monitored sensor, the physical state comprising one or more of position, connection, power-on, signal reception, or data transmission of the monitored sensor;

monitoring a data transmission state of the monitored sensor, the data transmission state comprising one or more of data size, format, frequency, resolution hash value, timestamp, reflection value, or height value of the monitored sensor;

monitoring output data of the monitored sensor, and using predetermined data to perform cross-validation for the output data;

when (a) a result of monitoring the physical state or the data transmission state indicates a deviation from a predefined range of values for one or more criteria comprised by the physical state or the data transmission state or (b) there is a validation result abnormality, determining that the monitored sensor is getting abnormal, and giving an alarm in a manner comprising at least one of an in-vehicle speech, an image prompt, or a remote notification; and

responsive to determining that the monitored sensor is getting abnormal, recording current traffic environment conditions and key data, and activating a safety response policy of the driverless vehicle;

wherein using the predetermined data to perform the cross-validation for the output data comprises performing cross-validation for the output data transmitted from the monitored sensor by using one of output data of other sensors other than the monitored sensor, a high-precision map, and data previously output by the monitored sensor, or combinations of at least two thereof;

wherein when the monitored sensor is a laser radar sensor, performing the cross-validation for output data of the monitored sensor comprises:

acquiring a point cloud recognition result obtained according to output data of the laser radar sensor;

when a vehicle or an obstacle on a road is recognized from the point cloud recognition result but the vehicle or the obstacle is not recognized from both an image recognition result obtained according to output data of an image sensor and an ultrasonic recognition result obtained according to output data of an ultrasonic sensor, determining the laser radar sensor as getting abnormal;

wherein the abnormity of the laser radar sensor is caused by a hacker's attack or interfering in detection by the laser radar sensor with a strong laser, wherein the hacker's attack includes attacking the laser radar sensor by absorbing waves emitted by the laser radar for concealing and disguising.

8. The non-transitory computer-readable storage medium according to claim 7 , wherein when the monitored sensor is a positioning sensor, performing the cross-validation for the output data of the monitored sensor comprises:

determining a lane where a positioning result output by the positioning sensor lies, according to the high-precision map;

acquiring an image recognition result obtained according to output data of an image sensor, and determining a lane where the driverless vehicle lies according to the image recognition result;

determining the positioning sensor as getting abnormal if the lane where the positioning result lies is inconsistent with the lane where the driverless vehicle lies;

or,

comparing the positioning result output by the positioning sensor with a positioning result previously output by the positioning sensor;

determining the positioning sensor as getting abnormal if a difference of the two positioning results does not match the travel speed of the driverless vehicle.

9. The non-transitory computer-readable storage medium according to claim 8 , wherein when the monitored sensor is an image sensor, performing the cross-validation for the output data of the monitored sensor comprises:

acquiring an image recognition result obtained according to output data of the image sensor;

when a traffic light or traffic sign is recognized from the image recognition result but it is determined by querying the high-precision map that the traffic light or traffic sign does not exist at a position where the driverless vehicle lies, determining the image sensor as getting abnormal;

or,

acquiring an image recognition result obtained according to output data of the image sensor;

when a vehicle or pedestrian is recognized from the image recognition result but the vehicle or pedestrian is not recognized from a point cloud recognition result obtained according to output data of a laser radar sensor, determining the image sensor as getting abnormal.