Work

A Deep Learning Framework for Skeleton-Based Gender Recognition from Drone Videos

Year: 2026

Type: conference-paper

Abstract: This study describes a deep learning model to perform gender recognition based on the skeleton in using gait video captured by a drone. Important A custom dataset of 4680 skeleton images was prepared by extracting 33 pose landmarks per frame with MediaPipe Pose and transforming them into clean stickman representations. An EfficientNet-B0 model with finetuning, using data augmentation, label smoothing and cosine annealing reached a 92% classification accuracy, which is far better than the performance of baseline CNN, MobileNetV3, ResNet18, and DenseNet121 models. Although aerial perspective and motion hiccups, as well as outdoor light fluctuation, are problematic to the proposed approach, overall the generalization and strength are high. These results illustrate how the system can be used in the real world to perform surveillance, crowd analytics, and biometric intelligence that avoids privacy invasion.

Source: International Conference on Power, Electronics, Communications, Computing, and Intelligent Infrastructure (PECCII)

DOI: 10.1109/peccii70991.2026.11661974

Authors: Mohammad Ali Sakib, Rafiu Mollick, Redwan Hussain, Zillur Rahaman, Abdullah Al Mahim, Nakib Aman

Institutions: Pabna University of Science and Technology, Daffodil International University, Bangladesh University of Business and Technology

Language: English

Topic: Gait Recognition and Analysis

Subfield: Biomedical Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Gender equality

Open Access status: hybrid

OpenAlex ID: W7204459610

FWCI:

Cites: 15

Cited by:

Related to: 10