摘 要
随着智能交通系统和自动驾驶技术的快速发展，车辆与行人的精准检测与识别成为计算机视觉领域的关键研究课题。本研究旨在提出一种基于深度学习的高效车辆行人检测与识别算法，以应对复杂场景下的目标识别挑战。为此，设计了一种改进的卷积神经网络（CNN）架构，通过引入多尺度特征融合机制和注意力机制，显著增强了对小目标和遮挡目标的检测能力。同时，采用数据增强技术和自适应锚框策略优化训练过程，进一步提升了模型在不同光照、天气条件下的鲁棒性。实验结果表明，该算法在公开数据集KITTI和COCO上的平均精度均值（mAP）分别达到了91.3%和87.6%，相较于现有主流方法具有明显优势。此外，算法在嵌入式设备上的实时性测试中表现出色，能够满足实际应用场景的需求。本研究的主要贡献在于提出了创新性的网络结构和优化策略，有效解决了复杂场景下目标检测的精度与效率问题，为智能交通系统和自动驾驶技术的发展提供了有力支持。
关键词：深度学习；车辆行人检测；卷积神经网络；多尺度特征融合；注意力机制
Abstract
With the rapid development of intelligent transportation systems and autonomous driving technologies, the accurate detection and recognition of vehicles and pedestrians have become key research topics in the field of computer vision. This study aims to propose an efficient deep-learning-based algorithm for vehicle and pedestrian detection and recognition, addressing the challenges of ob ject identification in complex scenarios. To achieve this, an improved convolutional neural network (CNN) architecture was designed, incorporating a multi-scale feature fusion mechanism and attention mechanism, which significantly enhances the detection capabilities for small and occluded ob jects. Additionally, data augmentation techniques and an adaptive anchor box strategy were employed to optimize the training process, further improving the model's robustness under varying lighting and weather conditions. Experimental results demonstrate that the proposed algorithm achieves mean average precision (mAP) values of 91.3% and 87.6% on the public datasets KITTI and COCO, respectively, showing a clear advantage over existing mainstream methods. Moreover, the algorithm performs excellently in real-time tests on embedded devices, meeting the requirements of practical application scenarios. The primary contribution of this study lies in the proposal of innovative network structures and optimization strategies, effectively resolving the issues of accuracy and efficiency in ob ject detection under complex conditions, thereby providing strong support for the development of intelligent transportation systems and autonomous driving technologies.
Keywords: Deep Learning；Vehicle Pedestrian Detection；Convolutional Neural Network；Multi-Scale Feature Fusion；Attention Mechanism
目 录
摘 要 I
Abstract II
一、绪论 1
（一）车辆行人检测与识别的研究背景 1
（二）深度学习在该领域的研究现状 1
（三）本文研究方法与技术路线 2
二、深度学习模型的构建与优化 2
（一）深度学习模型的选择与分析 2
（二）数据集构建与预处理方法 3
（三）模型优化策略与算法改进 3
（四）实验验证与性能评估 4
三、车辆检测算法的设计与实现 4
（一）车辆特征提取与表示方法 4
（二）基于深度学习的车辆检测框架 5
（三）算法复杂度与实时性分析 5
（四）实际场景中的应用效果 6
四、行人识别算法的研究与应用 6
（一）行人特征提取与分类方法 6
（二）深度学习在行人识别中的优势 7
（三）算法鲁棒性与适应性分析 7
（四）多场景下的实验结果与讨论 8
结 论 8
致 谢 10
参考文献 11
