US 12,406,380 B2
Image processing apparatus, image processing system, image processing method, and non-transitory computer-readable medium storing image processing program therein
Takashi Shibata, Tokyo (JP); and Takuya Ogawa, Tokyo (JP)
Assigned to NEC CORPORATION, Tokyo (JP)
Appl. No. 17/800,255
Filed by NEC Corporation, Tokyo (JP)
PCT Filed Mar. 12, 2020, PCT No. PCT/JP2020/010853
§ 371(c)(1), (2) Date Aug. 17, 2022,
PCT Pub. No. WO2021/181612, PCT Pub. Date Sep. 16, 2021.
Prior Publication US 2023/0107372 A1, Apr. 6, 2023
Int. Cl. G06T 7/246 (2017.01); G06V 10/62 (2022.01); G06V 10/776 (2022.01)
CPC G06T 7/248 (2017.01) [G06V 10/62 (2022.01); G06V 10/776 (2022.01); G06T 2207/10016 (2013.01); G06T 2207/30241 (2013.01)] 5 Claims
OG exemplary drawing
 
1. An image processing apparatus comprising:
at least one memory storing instructions, and
at least one processor configured to execute the instructions to:
extract an image feature of an object from a first time-series image of a plurality of time-series images that also include a plurality of subsequent time-series images from a second time-series image through a last time-series image;
identify a type or attribute of the object using the image feature extracted from the first time-series image;
predict an object position of the object in the first time-series image;
track the object through the plurality of subsequent time-series images by, for each time-series image from the second time-series image through the last time-series image:
selecting, from a plurality of kinetic models, a kinetic model to predict an object position of the object in the time-series image, using the identified type or attribute of the object and the predicted object position of the object in a prior time-series image;
predicting the object position of the object in the time-series image using the selected kinetic model;
predict the object position in each subsequent time-series image from a position of the object in a previous time-series image using the selected kinetic model;
generate a candidate for a trajectory of the object as a hypothesis from the predicted object position of the object;
calculate an object reliability of the object in an identification result;
calculate a movement reliability of a distance between the predicted position of the object and a detected position of the object in each time-series image;
calculate a reliability of the hypothesis by integrating the object reliability and the movement reliability; and
accumulate the reliability of the hypothesis in each time-series image and select the hypothesis having a highest accumulated cumulative reliability from a plurality of hypotheses as a tracking result.