Improved RepVGG-based Anchor-free Algorithm for On-road Object Detection
- Resource Type
- Conference
- Authors
- Lian, Zheng; Nie, Yiming; Dai, Bin; Xu, Xiaoyu
- Source
- 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC) Intelligent Transportation Systems (ITSC), 2022 IEEE 25th International Conference on. :1082-1087 Oct, 2022
- Subject
- Communication, Networking and Broadcast Technologies
Components, Circuits, Devices and Systems
Computing and Processing
Engineering Profession
Power, Energy and Industry Applications
Robotics and Control Systems
Signal Processing and Analysis
Transportation
Adaptive systems
Object detection
Feature extraction
Real-time systems
Inference algorithms
Task analysis
Intelligent transportation systems
- Language
On-road object detection is an important part of driverless technology. The on-road object detection task requires both detection speed and accuracy. We propose an improved RepVGG-based anchor-free real-time object detection algorithm to meet these requirements. The RepVggmodule is improved by a reparameterization method, and an adaptive Fusion-Distribution Feature Pyramid Network(FDFPN) structure is proposed, based on which an anchor-free object detection head with fewer hyperparameters is constructed to balance accuracy and speed. Experiments on KITTI dataset show that the accuracy of this method can reach 80.01%, and the inference latency is only 5.9ms in deployment mode.