US 11,915,099 B2
Information processing method, information processing apparatus, and recording medium for selecting sensing data serving as learning data
Takuya Yamaguchi, Osaka (JP)
Assigned to PANASONIC INTELLECTUAL PROPERTY CORPORATION OF AMERICA, Torrance, CA (US)
Filed by Panasonic Intellectual Property Corporation of America, Torrance, CA (US)
Filed on Jul. 18, 2019, as Appl. No. 16/515,385.
Claims priority of application No. 2018-145357 (JP), filed on Aug. 1, 2018.
Prior Publication US 2020/0042803 A1, Feb. 6, 2020
Int. Cl. G06K 9/00 (2006.01); G06T 7/70 (2017.01); G06V 20/56 (2022.01); G06V 20/58 (2022.01); G06V 40/10 (2022.01)
CPC G06V 20/58 (2022.01) [G06T 7/70 (2017.01); G06V 20/584 (2022.01); G06V 20/588 (2022.01); G06V 40/103 (2022.01); G06T 2207/30196 (2013.01); G06T 2207/30252 (2013.01)] 15 Claims
OG exemplary drawing
 
1. An information processing method, comprising:
executing an object detection process using an object detection model to which sensing data from a first sensor that detects an environment of a moving body is input, wherein the object detection model is a model which has been trained by a machine learning;
acquiring (i) a first object detection result obtained by the object detection process using the object detection model and (ii) a second object detection result obtained by use of a second sensor that detects an environment of the moving body;
specifying a specific region included in a third sensing space where a first sensing space of the first sensor overlaps a second sensing space of the second sensor;
determining a degree of agreement between the first object detection result and the second object detection result in the specific region specified;
selecting the sensing data as learning data to further train the object detection model by the machine learning in a case where the first object detection result does not agree with the second object detection result or in a case where the degree of agreement is lower than a predetermined value; and
training the object detection model using the sensing data selected as the learning data,
wherein the first sensor is a camera installed in the moving body,
the first sensing space is a capturing range of the camera,
the sensing data is an image of the environment of the moving body within the capturing range, the image being captured by the camera,
the executing of the object detection process using the object detection model is detecting (i) an object in the image and (ii) a region where the object in the image is located,
the first object detection result includes a first information piece related to the region of the detected object in the image,
the second sensor is a distance sensor installed in the moving body,
the second sensing space is a detection range of the distance sensor,
the second object detection result includes a second information piece including a three-dimensional shape of a terrain including an object in an environment of the moving body within the detection range, the three-dimensional shape being detected by the distance sensor,
the specific region is a region corresponding to a road or a sidewalk in the third sensing space,
when the specific region is the road, an object to be detected by use of the object detection model is a vehicle,
when the specific region is the sidewalk, the object to be detected by use of the object detection model is a person,
the first object detection result includes a first region which the object to be detected occupies in the specific region,
the second object detection result includes a second region which the object to be detected occupies in the specific region, and
in the determining, a degree of overlap to which the first region overlaps the second region is determined as the degree of agreement.