An accurate shared bicycle detection network based on faster R‐CNN

Abstract Detecting shared bicycles is an essential and challenging task. Deep learning has been widely used in object detection tasks urban scenes, such as vehicle detection. However, deep algorithms still face many difficulties challenges bicycle For example, the problem of large deformation small targets because camera far away from bicycles. In order to solve these problems, this study introduces feature fusion module deformable convolution into network, which improves efficiency This proposes enhanced faster R‐CNN network (A classic two‐stage network) for a dataset (SBD) constructed model training testing. Compared with original R‐CNN, mean average precision (mAP) method on SBD improved by 13%, indicates that provided more suitable detecting also conducts experiments Microsoft Common Objects (COCO) dataset, where achieves 40.2% mAP, 5.8% higher than before improvement.