FII-CenterNet: An Anchor-Free Detector With Foreground Attention for Traffic Object Detection
Siqi Fan, Fenghua Zhu, Shichao Chen, Hui Zhang, Bin Tian, Yisheng Lv, Fei–Yue Wang
Abstract
Most successful object detectors are anchor-based, which is difficult to adapt to the diversity of traffic objects. In this paper, we propose a novel anchor-free method, called FII-CenterNet, which introduces the foreground information to eliminate the interference of the complex background information in traffic scenes. The foreground region proposal network segments the foreground based on boxes-induced segmentation annotation, and midground is proposed to provide rich edge information of the objects. In addition to foreground location, scale information is also introduced to improve the regression performance. Extensive experimental results on two public datasets verify the benefits of the introduction of the foreground information, and demonstrate that our FII-CenterNet achieves the state-of-the-art performance in both accuracy and efficiency.