基于深度学习卷积神经网络模型的行人检测设计

Pedestrian detection design based on deep learning of convolutional neural network model

随着近年来5G与人工智能的崛起,无人驾驶技术在不断突破,行人检测作为无人驾驶中的重要任务之一,是一个具有重要研究意义的课题。文中采用深度学习框架Pytorch和目标检测网络YOLO进行行人检测,分别搭建了YOLO v3、YOLO v3轻量版YOLOv3-Tiny、YOLO v3与SPP-Net融合版本YOLOv3-SPP行人检测平台,并对小型目标和大型目标识别进行了详细的测试对比。测试结果表明,YOLO v3和YOLOv3-SPP平均置信度较高;YOLOv3-Tiny实时性较高,适用于计算量小的行人识别场景。

With the rise of 5G and artificial intelligence in recent years,unmanned driving technology is constantly making breakthrough.Pedestrian detection is one of the important tasks of unmanned driving,and it is an important research topic.This project will use deep learning framework Pytorch and target detection network YOLO for pedestrian detection,YOLO v3,YOLO v3 lightweight version YOLOv3-Tiny,YOLO v3 and SPP-Net fusion version YOLOv3-SPP are establised to test and being compared in terms of small and large targets.Experiment results show that YOLO v3 and YOLO v3-SPP has higher average confidence,while YOLOv3-Tiny has a low real-time performance which suits the pedestrian recognization situation of a small amount of calculation.