A Lightweight Multi-Modal Fusion Framework for In-Vehicle Occupant Detection on Embedded Systems.

Saved in:

Bibliographic Details
Title:	A Lightweight Multi-Modal Fusion Framework for In-Vehicle Occupant Detection on Embedded Systems.
Authors:	Wu, Enqi¹ wuenqi@usst.edu.cn, Liu, Qi² 15907957016@163.com, Pan, Yizhou² 1944520474@qq.com, Kou, Jiaming² bzmh007@163.com, Duan, Hanzuo² a1245481661@qq.com
Source:	IAENG International Journal of Computer Science. Aug2026, Vol. 53 Issue 8, p3154-3162. 9p.
Subjects:	Embedded computer systems, Automobile safety, Human facial recognition software, Object recognition (Computer vision), Multisensor data fusion
Abstract:	The unintentional abandonment of children in locked vehicles is a critical public safety issue often leading to fatal vehicular heatstroke. To mitigate this risk, this study presents a resource-efficient, multi-modal detection framework designed for embedded automotive platforms. Unlike traditional sensor-based methods prone to environmental noise, or heavy deep-learning models unsuitable for edge deployment, the proposed system integrates lightweight face classification with robust human body detection. Specifically, a cascaded architecture combines MTCNN for face localization with a structurally pruned Convolutional Neural Network (CNN) for age classification (child vs. adult). The pruning strategy reduces computational complexity by 73.7% while maintaining 99.4% accuracy, ensuring low-latency processing. For body detection, an optimized Single Shot MultiBox Detector (SSD) with a ResNet50 backbone is enhanced by Multi-scale Feature Fusion and an Efficient Channel Attention (ECA) mechanism to address occlusion challenges in vehicle cabins. Experimental results on the NVIDIA Jetson Nano platform demonstrate that the system achieves an Average Precision (AP) of 78.16% and operates at 16 FPS. This balance of high accuracy and computational efficiency provides a practical, real-time solution for preventing left-behind occupant incidents in intelligent vehicles. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Computer Science is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	The unintentional abandonment of children in locked vehicles is a critical public safety issue often leading to fatal vehicular heatstroke. To mitigate this risk, this study presents a resource-efficient, multi-modal detection framework designed for embedded automotive platforms. Unlike traditional sensor-based methods prone to environmental noise, or heavy deep-learning models unsuitable for edge deployment, the proposed system integrates lightweight face classification with robust human body detection. Specifically, a cascaded architecture combines MTCNN for face localization with a structurally pruned Convolutional Neural Network (CNN) for age classification (child vs. adult). The pruning strategy reduces computational complexity by 73.7% while maintaining 99.4% accuracy, ensuring low-latency processing. For body detection, an optimized Single Shot MultiBox Detector (SSD) with a ResNet50 backbone is enhanced by Multi-scale Feature Fusion and an Efficient Channel Attention (ECA) mechanism to address occlusion challenges in vehicle cabins. Experimental results on the NVIDIA Jetson Nano platform demonstrate that the system achieves an Average Precision (AP) of 78.16% and operates at 16 FPS. This balance of high accuracy and computational efficiency provides a practical, real-time solution for preventing left-behind occupant incidents in intelligent vehicles. [ABSTRACT FROM AUTHOR]
ISSN:	1819656X