Real-Time Threaded Houbara Detection and Segmentation for Wildlife Conservation

Lyes Saad Saoud, Loïc Lesobre, Enrico Sorato, Irfan Hussain

Khalifa University of Science and Technology, UAE
RENECO International Wildlife Consultants LLC, UAE
Preprint 2025

Paper Code
Houbara Detection System

The proposed system integrates YOLOv10 and MobileSAM for real-time detection and segmentation of Houbara Bustards in diverse environments.

Demo Videos

Without using TDM
Using TDM

Abstract

Real-time animal detection and segmentation play a crucial role in wildlife conservation. Our approach integrates YOLOv10 for object detection and MobileSAM for segmentation, with a threading-based model for improved efficiency. The system achieves high accuracy (mAP50 = 0.9627, mIoU = 0.7421) while enabling real-time performance on mobile devices. We introduce a curated dataset of 40,000 images for conservation AI applications.

Real-Time Houbara Analysis

(a) Input images (first, fourth, and seventh rows). (b) Predicted instance segmentation masks by MobileSAM (second, fifth, and eighth rows). (c) Combined view: Input images with object detections (bounding boxes) and MobileSAM segmentation masks (third, sixth, and ninth rows).

Real-time Houbara Analysis

Qualitative Comparison of Segmentation Results

This figure presents a qualitative comparison of segmentation performance across various models. Each column displays: (a) Input Images, (b) Ground Truth Annotations, (c) DeepLabV3 Results, (d) FCN Results, (e) LRASPP Results, (f) MobileSAM Baseline Results, (g) YOLOv9+MobileSAM Results, and (h) Overlapping Results of YOLOv10+MobileSAM with Input Images. The overlapping panel highlights the superior object localization and segmentation precision achieved by YOLOv10+MobileSAM.

Qualitative Segmentation Comparison

📖 BibTeX

@article{SaadSaoud2025HoubaraDetection,
  author    = {Lyes Saad Saoud and Loïc Lesobre and Enrico Sorato and Irfan Hussain},
  title     = {Real-Time Threaded Houbara Detection and Segmentation for Wildlife Conservation Using Mobile Platforms},
  journal   = {Preprint},
  year      = {2025},
  publisher = {arXiv},
  url       = {https://arxiv.org/abs/XXXX.XXXXX}
}