To reduce processing cost while suppressing accuracy deterioration resulting from a difference in environment during acquiring an image in the detection of an object from the image.

A hypothesis generating part 15 associates types defined in a detection candidate type table 13 with each detection candidate extracted from an image of a detection object, and generates a plurality of combinations of types about each detection candidate of the image as a hypothesis. An evaluating part 17 gives an evaluation point set in an evaluation point table 16 to a hypothesis having a predetermined detection inter-candidate condition to evaluate a hypothesis. A detecting part 18 specifies the type of each detection candidate by selecting one of hypotheses on the basis of an evaluation result, and detects a detection object on the basis of the specified type. A hypothesis generation control part 21 controls the hypothesis generating part 15 on the basis of an ambient environment of the detection object during photographing the image to restrict the generation of a hypothesis that is estimated not to occur in the ambient environment.