Smart Application for Real Time Detection: An Improved Lightweight Detector for Real-Time Vehicle Detection
Source of Publication
2023 Fourteenth International Conference on Ubiquitous and Future Networks (ICUFN)
With the goal of creating a model that satisfies the demands of detecting vehicles on the road in real time and delivers high FPS speed and high accuracy, we propose enhancements to the YOLOv4-tiny detector in this work. We suggest a modification to the FPN to increase the semantic and location information between the higher and lower levels. Also, to further solve the issue with NMS, we employed soft-NMS in our model. Moreover, to address the drawbacks of the IoU, we proposed a way to improve anchor clustering. The experimental results show that our proposed model outperforms the basic model with 3.29% higher mAP and 1.17 higher FPS.
Location awareness, Roads, Semantics, Road vehicles, Detectors, Predictive models, Real-time systems
Alsanabani, Ala; Abugabah, Ahed; and Jiao, Licheng, "Smart Application for Real Time Detection: An Improved Lightweight Detector for Real-Time Vehicle Detection" (2023). All Works. 5970.
Indexed in Scopus