Object detection device, method, and program

An object detection method comprising:

• by a computer,
• calculating a motion vector of a target frame, wherein the target frame represents a current frame of a series of frames of images collectively describing a moving image, the moving image depicts an object in motion, the motion vector indicates change of a location of the object in the target frame in pixels as the object having moved from a previous location in a reference frame to a current location in the current frame, and the reference frame represents a preceding frame that precedes the current frame in the series of frames of images;
• calculating, based on the motion vector, an index indicating a magnitude of change of locations of the object between a key frame and the target frame, wherein the key frame represents the closest previous frame of the series of frames where the object was detected by processing an object detection model;
• when the index is equal to or greater than a predetermined threshold value, detecting a first region indicating the object in the target frame by using the object detection model;
• when the index is less than the predetermined threshold value, detecting a second region indicating the object in the target frame by correcting a position of the location of the object detected in the reference frame by using according to the motion vector;
• and
• outputting, selectively based on first information of the first region and second information of the second region according to the index, an updated location of the object in the target frame of the series of frames of images with accuracy.

A non-transitory computer readable medium storing a program executable by a computer to perform a process for object detection processing, the process comprising:

• calculating a motion vector of a target frame, wherein the target frame represents a current frame of a series of frames of images collectively describing a moving image, the moving image depicts an object in motion, the motion vector indicates change of a location of the object in the target frame in pixels as the object having moved from a previous location in a reference frame to a current location in the current frame, and the reference frame represents a preceding frame that precedes the current frame in the series of frames of images;
• calculating, based on the motion vector, an index indicating a magnitude of change of locations of the object between a key frame and the target frame, wherein the key frame represents the closest previous frame of the series of frames where the object was detected by processing an object detection model;
• when the index is equal to or greater than a predetermined threshold value, detecting a first region indicating the object in the target frame by using the object detection model;
• when the index is less than the predetermined threshold value, detecting a second region indicating the object in the target frame by correcting a position the location of the object detected in the reference frame according to the motion vector;
• and
• outputting, selectively based on first information of the first region and second information of the second region according to the index, an updated location of the object in the target frame of the series of frames of images with accuracy.