Object detecting device and method

1 . An object detecting device comprising:

one or more processors; and

a non-transitory storage device storing a program to be executed by the one or more processors, the program including instructions to:

divide an input image into a plurality of segmented images;

perform object detection processing on a reduced image obtained by scaling a whole of the input image for each frame of the input image to obtain an overall object detection result;

determine a first segmented image for each frame of one or more frames which is subject to object detection among the plurality of segmented images in advance in a cycle of N frames, wherein N is an integer equal to or greater than a total number of segmented images, and reserve the first segmented image to obtain a reservation result;

select and output, for each frame of the one or more frames, one or more selected segmented images that are subject to object detection among the plurality of segmented images based on the reservation result;

perform object detection processing on each of one or more selected segmented images to obtain a division object detection result; and

obtain a final object detection result by synthesizing the overall object detection result and the division object detection result,

wherein the instructions to determine a segmented image for each frame of the plurality of segmented images which is subject to object detection among the plurality of segmented images in advance in the cycle of N frames comprises instructions to:

calculate an estimated value of a quantity of detected objects for each of the plurality of segmented images for each frame of the one or more frames based on the overall object detection result;

calculate an estimated value of a fluctuation between frames of the quantity of detected objects for each of the plurality of segmented images based on an absolute difference value between the frames of the estimated value;

calculate an average value of estimated values of fluctuations of the quantity of detected objects within the cycle for each of the plurality of segmented images; and

reserve a segmented image for each frame which is subject to object detection in a next cycle such that a quantity of times object detection is executed is proportionally distributed to each segmented image based on an average value of estimated values of fluctuations of the quantity of detected objects for each segmented image.

2 . The object detecting device according to claim 1 , wherein the instructions include further instructions to:

after determining that all segmented images are selected for object detection at least M times in a next cycle, wherein M is an integer of 1 or more, determine a segmented image for each frame which is subject to object detection in the next cycle such that the number of times object detection is executed is proportionally distributed to each segmented image based on an average value of estimated values of fluctuations of the number of detected objects for each of the segmented images for the remaining allocable frames.

3 . The object detecting device according to claim 2 , wherein the instructions include further instructions to:

calculate the estimated value of the quantity of detected objects for each of the plurality of segmented images by aggregating a quantity of detected objects of which reliability calculated is less than a threshold value and equal to or greater than a threshold value/α for each segmented image in which the object is positioned, within an object group detected in the overall object detection result, wherein α is a real number of 2 or more.

4 . The object detecting device according to claim 2 , wherein the instructions include further instructions to:

store the division object detection result; and

select and output, for each frame of the one or more frames, a past object detection result for a second segmented image for which object detection is not executed based on the reservation result, wherein the overall object detection result, the division object detection result, and the past object detection result output for each frame of the one or more frames to obtain the final object detection result.

5 . The object detecting device according to claim 2 , wherein the instructions include further instructions to:

determine whether the input image has changed based on a latest object detection result and a past object detection result; and

cancel and returns the reservation to an initial state when it is determined that the input image has changed.

6 . The object detecting device according to claim 1 , wherein the instructions include further instructions to:

calculate the estimated value of the quantity of detected objects for each of the plurality of segmented images by aggregating a quantity of detected objects of which reliability calculated is less than a threshold value and equal to or greater than a threshold value/α for each segmented image in which the object is positioned, within an object group detected in the overall object detection result, wherein α is a real number of 2 or more.

7 . The object detecting device according to claim 1 , wherein the instructions include further instructions to:

store the division object detection result; and

select and output, for each frame of the one or more frames, a past object detection result for a second segmented image for which object detection is not executed based on the reservation result, wherein the overall object detection result, the division object detection result, and the past object detection result output for each frame of the one or more frames to obtain the final object detection result.

8 . The object detecting device according to claim 1 , wherein the instructions include further instructions to:

determine whether the input image has changed based on a latest object detection result and a past object detection result; and

cancel and return the reservation to an initial state when it is determined that the input image has changed.

9 . An object detection method comprising:

a first step of dividing an input image into a plurality of segmented images;

a second step of performing object detection processing on a reduced image obtained by scaling a whole of the input image for each frame of one or more frames;

a third step of determining a segmented image for each frame of the one or more frames which is subject to object detection among the plurality of segmented images in advance in a cycle of N frames, wherein N is an integer equal to or greater than a total number of segmented images, and reserving the segmented image;

a fourth step of selecting and outputting, for each frame, one or more first segmented images that are subject to object detection among the plurality of segmented images based on a reservation result from the third step;

a fifth step of performing object detection processing on each of the one or more first segmented images output in the fourth step; and

a sixth step of obtaining a final object detection result by synthesizing, for each frame, an object detection result from the second step and an object detection result from the fifth step,

wherein the third step comprises:

calculate an estimated value of a quantity of detected objects for each of the plurality of segmented images for each frame of the one or more frames based on the object detection result from the second step;

calculate an estimated value of a fluctuation between frames of the quantity of detected objects for each of the plurality of segmented images based on an absolute difference value between the frames of the estimated value;

calculate an average value of estimated values of fluctuations of the quantity of detected objects within the cycle for each of the plurality of segmented images; and

reserve a segmented image for each frame which is subject to object detection in a next cycle such that a quantity of times object detection is executed is proportionally distributed to each segmented image based on an average value of estimated values of fluctuations of the quantity of detected objects for each segmented image.

10 . The object detecting method according to claim 9 , further comprising:

a seventh step of after determining that all segmented images are selected for object detection at least M times in a next cycle, wherein M is an integer of 1 or more, determining a segmented image for each frame which is subject to object detection in the next cycle such that the number of times object detection is executed is proportionally distributed to each segmented image based on an average value of estimated values of fluctuations of the number of detected objects for each of the segmented images for the remaining allocable frames.

11 . An object detecting device comprising:

one or more processors; and

a non-transitory storage device storing a program to be executed by the one or more processors, the program including instructions to:

divide an input image into a plurality of segmented images;

perform object detection processing on a reduced image obtained by scaling a whole of the input image for each frame of the input image to obtain an overall object detection result;

determine a first segmented image for each frame of one or more frames which is subject to object detection among the plurality of segmented images in advance in a cycle of N frames, wherein N is an integer equal to or greater than a total number of segmented images, and reserve the first segmented image to obtain a reservation result;

select and output, for each frame of the one or more frames, one or more selected segmented images that are subject to object detection among the plurality of segmented images based on the reservation result;

perform object detection processing on each of one or more selected segmented images to obtain a division object detection result;

store the division object detection result;

select and output, for each frame of the one or more frames, a past object detection result for a second segmented image for which object detection is not executed based on the reservation result among the plurality of segmented images; and

obtain a final object detection result by synthesizing the overall object detection result, the division object detection result, and the past object detection result output for each frame of the one or more frames.