Computer software module arrangement, a circuitry arrangement, an arrangement and a method for improved object detection adapting the detection through shifting the image

A method for object detection utilizing a multi-scale convolutional neural network (CNN) that, with respect to detecting objects, exhibits dips in object classification confidence at repeating locations within an image coordinate system, wherein the method comprises:

• receiving an image comprising image data in which an object to be detected is depicted;
• classifying whether the depicted object is at risk of being incorrectly detected by the multi-scale CNN, wherein the risk is associated with a position of the depicted object relative to the repeating locations;
• and
• responsive to classifying the depicted object as being at risk of being incorrectly detected by the multi-scale CNN, compensating detection of the depicted object by the multi-scale CNN by shifting the image data within the image coordinated system.

A non-transitory computer-readable medium carrying computer instructions that when loaded into and executed by a controller of an object detection arrangement configures the object detection apparatus to:

• receive an image comprising image data in which an object to be detected by a multi-scale convolutional neural network (CNN) is depicted, and wherein, with respect to detecting objects, the multi-scale CNN exhibits dips in object classification confidence at repeating locations within an image coordinate system;
• classify whether the depicted object is at risk of being incorrectly detected by the multi-scale CNN, wherein the risk is associated with a position of the depicted object relative to the repeating locations;
• and
• responsive to classifying the depicted object as being at risk of being incorrectly detected by the multi-scale CNN, compensating detection of the depicted object by the multi-scale CNN by shifting the image data within the image coordinate system.