Abstract
AbstractObject detection of vulnerable road users (VRU) under low computing resources of roadside units is one of the key technologies to achieve vehicle-infrastructure cooperative perception. In this paper, a lightweight fine-grained VRU detection model is proposed. Analyzing the existing complex traffic environment, the traditional definition of VRU is no longer applicable. Our work includes two parts: One is to redefine the fine-grained VRU and construct a new dataset. This task makes the perceptual information obtained by detection more comprehensive and accurate. Another is to optimize YOLOv4 by using the channel pruning method in model compression. The optimized model is 60% lighter than the original model. Under the limitation of low computing resources at the roadside units, the real-time detection of VRU is realized while ensuring a certain detection accuracy.
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
