Abstract

Aiming at the problem that YOLOv5 has a high missed detection rate in the task of pedestrian detection in complex scenes, the ECA-YOLOv5 pedestrian detection algorithm is proposed. Aiming at the problems of high missed detection rate and complex scene in the complex crowded pedestrian detection task, the original BottleneckCSP module is replaced with the C3TR module with TransformerBlock, which has better performance in the detection of high-density occluded objects in the case of pedestrian detection. Aiming at the problem of excessive parameters such as redundant expansion of model training calculation, a lightweight general upsampling operator CARAFE is introduced. Compared with other upsampling methods, it can not only achieve better performance in multi-scene tasks, but also greatly reduce The parameter calculation amount is reduced, the detection speed is improved while the detection accuracy is preserved. The self-designed efficientCoordAtt high-efficiency attention mechanism module is used to enhance the receptive field and the model's ability to accurately locate the target, improve the detection accuracy in various complex scenes, and strengthen the model's ability to capture local information. The experimental results show that the improved ECA-YOLOv5s algorithm can effectively improve the mAP of pedestrian detection while maintaining the high real-time performance of the original algorithm.

Full Text
Published version (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