The perception module plays a crucial role in the autonomous driving system, which is a very complicated system. In contrast, current perception research in the area of autonomous driving primarily focuses on the recognition of vehicles, lanes, and traffic signs, ignoring other factors that may cause traffic accidents, and fails to consider the fact that many traffic accidents on highways are caused by wild or wandering caused by animals. We conducted the following studies to close this gap in knowledge: First, a dataset of 1050 photos for huge animals that could appear on highways was created. Next, we propose a more efficient Yolo model by improving its backbone, replacing the C3 module with C3Ghost. The number of parameters is decreased to fewer than 3.7 million, just 52.7% of Yolov5s, but the average accuracy for each type of animals (mAP% 0.5) has reached over 95%. Third, our GhostSort- YoloNet (GS-YoloNet) also incorporates the Deep Sort algorithm to achieve real-time ranging and speed assessment of numerous targets, which has a very big practical application promise.
A Real-Time Large Animal Detection Lightweight Network for Autonomous Driving on Highways
27.10.2023
5376323 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch