Issuu

International Research Journal of Engineering and Technology (IRJET)

e-ISSN: 2395-0056

Volume: 11 Issue: 09 | Sep 2024

p-ISSN: 2395-0072

www.irjet.net

Pedestrian detection in adverse lighting conditions Abdul Muqtadeer Ahmed1, Dr. Jyothi S Nayak2 1Abdul Muqtadeer Ahmed, Student, Dept of Computer Science and Engineering, BMS College of Engineering,

Karnataka, India

2Dr. Jyothi S Nayak, HOD & Professor, Dept of Computer Science and Engineering, BMS College of Engineering,

Karnataka, India ---------------------------------------------------------------------***---------------------------------------------------------------------

Abstract - In this study, a new method based on YOLOv7, an

Even with the development of object detection architectures like YOLO (You Only Look Once), performance degradation in difficult lighting conditions is still a problem. For example, notable advances in real-time object detection have been made by YOLOv3 by Redmon et al. [1], YOLOv4 by Bochkovskiy et al. [2], and YOLOv5 by Glenn Jocher [3]. However, their performance is reduced in low-light conditions.

Our study expands upon earlier research, such as YOLOv3 by Redmon et al. [1] and YOLOv4 by Bochkovskiy et al. [2]. Additionally, it makes use of adaptive feature improvement, which was motivated by Chen et al. [3], and data augmentation techniques as explained by Zhang et al. [4].

In order to significantly improve pedestrian detection and tracking accuracy in low-light conditions, the YOLOv7 architecture will be modified using carefully chosen lowlight datasets. This research also seeks to track pedestrian paths, improve pedestrian detection efficiency, and give each pedestrian a unique identification.

advanced object detection model specially designed with customized datasets is proposed to detect and track pedestrian under low-light conditions. Pedestrian detection and tracking have been a great problem for years. Because it is a very important technique used in various applications such as person identification, surveillance system and autonomous vehicles.

After wide testing on a vast range of datasets that were handpicked to accurately reflect difficult low-light scenarios, the modified YOLOv7 model shows impressive flexibility when it comes to detecting and following pedestrians in the face of complex lighting changes. The efficiency of our suggested methodology is confirmed by quantitative assessments, which show prominent gains in precision and recall rates when compared to previous YOLO versions.

Previous research emphasizes how traditional pedestrian detection techniques are limited in difficult lighting conditions. YOLOv3 was first presented by Redmon et al. [1], who emphasized accuracy and speed in object detection tasks. The YOLO architecture in YOLOv4 was further optimized by Bochkovskiy et al. [2], improving performance metrics on a variety of datasets. Enhancements in training strategies and model scaling were introduced by Jocher's YOLOv5 [3]. However, none of these versions have sufficiently tackled the subtle problems caused by unfavorable lighting. By suggesting changes to the YOLOv7 architecture that are motivated by current developments in feature enhancement and data augmentation techniques, our paper seeks to close this crucial gap in this regard [4][5]. By means of comprehensive testing with a varied dataset that has been carefully selected to depict difficult low-light situations, the suggested method seeks to:

By highlighting the usefulness and significance of exploiting YOLOv7 to improve accuracy in difficult visual environments, this study significantly advances the field of pedestrian detection and tracking in low-light situations. Key Words: YOLOv7, Datasets, Accuracy, Precision, Mean Average Precision.

1.INTRODUCTION A crucial component of machine learning is pedestrian detection, which forms the basis for a variety of applications, from autonomous vehicles to surveillance systems. However, because of shadows, dim light, and varying illumination, low lighting conditions present a serious challenge to pedestrian detection algorithms' accuracy. These difficulties are especially noticeable in situations where autonomous vehicles that only have visual sensors find it difficult to identify moving objects on the road and avoid collisions.

Impact Factor value: 8.315

Improve pedestrian detection accuracy in adverse lighting conditions.

Increase detection efficiency.

Enable tracking of pedestrian paths.

Uniquely identify each pedestrian.

Our work paves the way for the development of more durable and dependable computer vision systems that can detect and track pedestrians efficiently in low-light conditions by tackling these goals.

ISO 9001:2008 Certified Journal

Page 878