Abstract:Real-time detection and tracking of pedestrian targets is a critical aspect of autonomous driving. However, due to complex road conditions, bad weather and other factors, traditional image detection algorithms are difficult to accurately detect pedestrians on the road, which leads to traffic accidents. To solve the above problems, this paper proposes an improved pedestrian detection and tracking algorithm using YOLOv8n and DeepSort. In the detection stage, the RC-YOLOv8n model is designed. Considering the high real-time requirement in driverless technology, to ensure that the system can quickly respond to a variety of complex road conditions and emergencies, the lightweight RepViT model is introduced to achieve efficient and accurate real-time detection. Meanwhile, the C2f-DCN (Deformable Convolutional Networks) module is proposed to improve the detection accuracy and speed of the model by using deformable convolution. In the tracking phase, the CA-DeepSort model is proposed. The model replaces the pedestrian re-recognition network in DeepSort with the CANet36 network, which captures the spatial location information of pedestrians more accurately by introducing the coordinate attention mechanism, thus improving the fineness and recognition of the feature representation. The CA coordinate attention mechanism is an attention mechanism that enhances the ability of convolutional neural networks to capture spatial and channel information. In addition, Wise-IoU is introduced to replace the traditional IoU for matching, which comprehensively takes into account the shape and position features of the target object, thus improving the matching accuracy. After conducting experiments on different datasets and analysing the experimental data, it is concluded that the RC-YOLOv8n algorithm improves the mAP by 1.6%. When combined with the CA-DeepSort algorithm for tracking, the MOTA reaches 73.2%, which is 12.8% higher than the original algorithm, and the MOTP is 81.9%, which is 1.4% higher than the original algorithm. These improvements significantly enhance the pedestrian detection and tracking performance of driverless vehicles.

Real-time Pedestrian Tracking Based on YOLOv3 and Prototype Clustering

Pedestrian Detection Method Based on Improved YOLOv5s for Densely Occluded Scenarios

Towards Accurate Dense Pedestrian Detection Via Occlusion-Prediction Aware Label Assignment and Hierarchical-Nms.

Real-Time Pedestrian Detection and Tracking Based on YOLOv3

Mobile Robot Tracking Method Based on Improved YOLOv8 Pedestrian Detection Algorithm

Real‐time 3D multi‐pedestrian detection and tracking using 3D LiDAR point cloud for mobile robot

An Efficient Pedestrian Detection for Realtime Surveillance Systems Based on Modified YOLOv3

Clustering based detection of small target pedestrians for smart cities

Dynamic Video Streaming Real-time Headcount Based on Improved YOLO V5 for Open Plazas

A Pedestrian Detection Network Model Based on Improved YOLOv5

Vision Based Multi-pedestrian Tracking Using Adaptive Detection and Clustering.

Research on Pedestrian Tracking Technology for Autonomous Driving Scenarios

Real-Time 3D Object Detection on Crowded Pedestrians

Research on Pedestrian Detection and DeepSort Tracking in Front of Intelligent Vehicle Based on Deep Learning

Lightweight Pedestrian Detection Network for UAV Remote Sensing Images Based on Strideless Pooling

Robust real-time pedestrian detection on embedded devices

Pedestrian Partial Occlusion Detection Method Based on Improved YOLO Network Structure

Multi-objective pedestrian tracking method based on YOLOv8 and improved DeepSORT

End-to-End Network for Pedestrian Detection, Tracking and Re-Identification in Real-Time Surveillance System

YOLOv8-based Dense Pedestrian Detection Algorithm

YOLOpeds: Efficient Real-Time Single-Shot Pedestrian Detection for Smart Camera Applications