Research on the object detection in complex scenes using diagonal distance

Abstract The detection method based on IoU boundary box is an important method of image recognition. This paper proposes a diagonal distance loss function, , and a combined diagonal distance loss function, , based IoU. These functions use the distance between diagonals as a penalty, overcoming the degradation phenomenon of prior functions based on traditional IoU. The altered loss functions are also integrated into faster R‐CNN and SSD models, and their effectiveness is assessed on PASCAL VOC, MS COCO and SSDD data sets, yielding favourable results. This paper also extends the IoU calculation between two boxes when performing non‐maximum suppression to advance position precision. The experiments have demonstrated that the enhanced distance calculation method improves both loss function and non‐maximum suppression calculations by around .