Related Experiment Video
Updated: Aug 22, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.0K
A Lightweight Vehicle-Pedestrian Detection Algorithm Based on Attention Mechanism in Traffic Scenarios
Yong Zhang1, Aibo Zhou1, Fengkui Zhao1
1School of Automotive and Traffic Engineering, Nanjing Forestry University, Nanjing 210037, China.
Sensors (Basel, Switzerland)
|November 11, 2022
Summary
This study introduces an improved YOLOv4 algorithm for faster and more accurate vehicle and pedestrian detection in autonomous driving. The enhanced model balances speed and accuracy, crucial for real-world traffic scenarios.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Autonomous Systems
Background:
- Object detection is vital for autonomous vehicle perception.
- Convolutional Neural Networks (CNNs) excel at feature extraction for vehicle detection.
- Balancing detection speed and accuracy in complex traffic is a key challenge.
Purpose of the Study:
- To develop a lightweight, high-performance vehicle-pedestrian detection algorithm.
- To improve upon the YOLOv4 model for enhanced efficiency and accuracy.
- To achieve a better balance between detection speed and accuracy in complex traffic scenarios.
Main Methods:
- Replaced the CSPDarknet53 backbone with MobileNetv2 for reduced parameters and improved feature extraction.
- Implemented multi-scale feature fusion for enhanced information interaction across feature layers.
- Incorporated a coordinate attention mechanism to focus on regions of interest via weight adjustment.
Main Results:
- The improved YOLOv4 model achieved a mean Average Precision (mAP) of 85.79% and a speed of 35 Frames Per Second (FPS) on PASCAL VOC datasets.
- Demonstrated a 4.31% increase in mAP and a 16.7% increase in speed compared to the original YOLOv4.
- The model shows a significant improvement in vehicle-pedestrian detection performance within traffic scenarios.
Conclusions:
- The proposed improved YOLOv4 algorithm offers a superior balance between detection accuracy and speed.
- This lightweight and high-performance model is suitable for real-world vehicle-pedestrian detection in autonomous driving applications.
- The enhancements contribute to more robust environmental perception for autonomous vehicles.
