Object detection device, object detection method, and object detection program

An object detection method comprising:

• using a predetermined prediction model to acquire, from a known first image for which the coordinates of a detection subject are known, a confidence value with respect to each detection region extracted from a plurality of positions in the first image and indicating certainty that the detection subject exists in the detection region;
• on the basis of the acquired confidence value, determining parameters that include a detection threshold value which is a threshold value with respect to the confidence value, said parameters being used when detecting the detection subject from a second image in which it is not known whether the detection subject exists;
• on the basis of the determined parameters, narrowing down detection region candidates to be used as detection region extraction sources from the entire region of a second image;
• using the prediction model to acquire the confidence value with respect to each detection region extracted from the narrowed-down detection region candidates;
• and
• detecting the detection subject on the basis of the acquired confidence value.

A non-transitory computer-readable recording medium having recorded therein an object detection program causing a computer to execute:

• first object detection processing of using a predetermined prediction model to acquire, from a known first image for which the coordinates of a detection subject are known, a confidence value with respect to each detection region extracted from a plurality of positions in the first image and indicating certainty that the detection subject exists in the detection region;
• parameter determination processing of, on the basis of the acquired confidence value, determining parameters that include a detection threshold value which is a threshold value with respect to the confidence value, said parameters being used when detecting the detection subject from a second image in which it is not known whether the detection subject exists;
• and
• second object detection processing of, on the basis of the determined parameters, narrowing down detection region candidates to be used as detection region extraction sources from the entire region of a second image, using the prediction model to acquire the confidence value with respect to each detection region extracted from the narrowed-down detection region candidates, and detecting the detection subject on the basis of the acquired confidence value.