BIoU: An Improved Bounding Box Regression for Object Detection

Object detection is a predominant challenge in computer vision and image processing to detect instances of objects various classes within an or video. Recently, new domain vehicular platforms, e-scooters, has been widely used across domestic urban environments. The driving behavior e-scooter users significantly differs from other vehicles on the road, their interactions with pedestrians are also increasing. To ensure pedestrian safety develop efficient traffic monitoring system, reliable object system for e-scooters required. However, existing detectors based IoU loss functions suffer drawbacks when dealing densely packed inaccurate predictions. address this problem, function, balanced-IoU (BIoU), proposed article. This function considers parameterized distance between centers minimum maximum edges bounding boxes localization problem. With help synthetic data, simulation experiment was carried out analyze box regression losses. Extensive experiments have two-stage detector, MASK_RCNN, single-stage such as YOLOv5n6, YOLOv5x Microsoft Common Objects Context, SKU110k, our custom dataset. demonstrated increment 3.70% at APS COCO dataset, 6.20% AP55 9.03% AP80