Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A lightweight YOLOv5-FFM model for occlusion pedestrian detection

Xiangjie Luo, Bo Shao, Zhihao Cai, Yingxun Wang

TL;DR

This paper tackles occlusion-aware pedestrian detection in autonomous driving by enhancing the lightweight YOLOv5s model with SE blocks, Ghost-based modules, a Wise-IoU loss, and a novel Feature Fusion Module (FFM). The FFM fuses local head and leg features into coherent full-body predictions to recover occluded pedestrians, while Ghost modules and SE blocks reduce computation and boost global context understanding. The approach yields significant AP gains on CityPersons and CUHK Occlusion with substantial parameter and FLOP reductions, bringing near-parallel performance to larger YOLOv5 variants at a fraction of their cost. This combination enables robust, real-time pedestrian detection in occluded scenarios, with clear paths for further refinement in pose variability and NMS handling.

Abstract

The development of autonomous driving technology must be inseparable from pedestrian detection. Because of the fast speed of the vehicle, the accuracy and real-time performance of the pedestrian detection algorithm are very important. YOLO, as an efficient and simple one-stage target detection method, is often used for pedestrian detection in various environments. However, this series of detectors face some challenges, such as excessive computation and undesirable detection rate when facing occluded pedestrians. In this paper, we propose an improved lightweight YOLOv5 model to deal with these problems. This model can achieve better pedestrian detection accuracy with fewer floating-point operations (FLOPs), especially for occluded targets. In order to achieve the above goals, we made improvements based on the YOLOv5 model framework and introduced Ghost module and SE block. Furthermore, we designed a local feature fusion module (FFM) to deal with occlusion in pedestrian detection. To verify the validity of our method, two datasets, Citypersons and CUHK Occlusion, were selected for the experiment. The experimental results show that, compared with the original yolov5s model, the average precision (AP) of our method is significantly improved, while the number of parameters is reduced by 27.9% and FLOPs are reduced by 19.0%.

A lightweight YOLOv5-FFM model for occlusion pedestrian detection

TL;DR

This paper tackles occlusion-aware pedestrian detection in autonomous driving by enhancing the lightweight YOLOv5s model with SE blocks, Ghost-based modules, a Wise-IoU loss, and a novel Feature Fusion Module (FFM). The FFM fuses local head and leg features into coherent full-body predictions to recover occluded pedestrians, while Ghost modules and SE blocks reduce computation and boost global context understanding. The approach yields significant AP gains on CityPersons and CUHK Occlusion with substantial parameter and FLOP reductions, bringing near-parallel performance to larger YOLOv5 variants at a fraction of their cost. This combination enables robust, real-time pedestrian detection in occluded scenarios, with clear paths for further refinement in pose variability and NMS handling.

Abstract

The development of autonomous driving technology must be inseparable from pedestrian detection. Because of the fast speed of the vehicle, the accuracy and real-time performance of the pedestrian detection algorithm are very important. YOLO, as an efficient and simple one-stage target detection method, is often used for pedestrian detection in various environments. However, this series of detectors face some challenges, such as excessive computation and undesirable detection rate when facing occluded pedestrians. In this paper, we propose an improved lightweight YOLOv5 model to deal with these problems. This model can achieve better pedestrian detection accuracy with fewer floating-point operations (FLOPs), especially for occluded targets. In order to achieve the above goals, we made improvements based on the YOLOv5 model framework and introduced Ghost module and SE block. Furthermore, we designed a local feature fusion module (FFM) to deal with occlusion in pedestrian detection. To verify the validity of our method, two datasets, Citypersons and CUHK Occlusion, were selected for the experiment. The experimental results show that, compared with the original yolov5s model, the average precision (AP) of our method is significantly improved, while the number of parameters is reduced by 27.9% and FLOPs are reduced by 19.0%.
Paper Structure (16 sections, 3 equations, 10 figures, 3 tables)

This paper contains 16 sections, 3 equations, 10 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Methodologies
  4. YOLOv5 model
  5. Overview of the proposed network
  6. Redesign of the YOLOv5s network architecture
  7. Improvement of the backbone network
  8. Improvement of the neck network
  9. Loss function improvement
  10. Feature fusion module(FFM)
  11. Experiments
  12. Ablation study
  13. Experiment on Citypersons dataset
  14. Experiment on CUHK Occlusion dataset
  15. Discussions
  16. ...and 1 more sections

Figures (10)

  • Figure 1: The structural diagram of the improved YOLOv5s with FFM
  • Figure 2: A squeeze-and-excitation block
  • Figure 3: The ordinary convolution operations
  • Figure 4: The Ghost module
  • Figure 5: GhostBottleneck and C3Ghost
  • ...and 5 more figures