International Research Journal of Engineering and Technology (IRJET) Volume: 12 Issue: 04 | Apr 2025
www.irjet.net
e-ISSN: 2395-0056 p-ISSN: 2395-0072
Camouflaged Object Detection System Using YOLOv8 Segmentation Prof. Swati Dronkar 1, Kunalsingh Bais2, Ishan Jaiswal3 Nikhil Khawase4, Yash Dipke5, Yash Channawar6 Prof. Swati Dronkar ,CSE , Priyadarshini College of Engineering Nagpur, Maharashtra, India Kunalsingh Bais , CSE ,Priyadarshini College of Engineering Nagpur, Maharashtra, India 3. Ishan Jaiswal ,CSE ,Priyadarshini College of Engineering Nagpur, Maharashtra, India 4. Nikhil Khawase ,CSE , Priyadarshini College of Engineering Nagpur, Maharashtra, India 5. Yash Dipke ,CSE , Priyadarshini College of Engineering Nagpur, Maharashtra, India 6. Yash Channawar ,CSE , Priyadarshini College of Engineering Nagpur, Maharashtra, India ---------------------------------------------------------------------***--------------------------------------------------------------------1.
2.
Abstract - In this paper, we introduce a novel
Our system architecture includes:
camouflaged object detection system that leverages the YOLOv8 segmentation model, fine-tuned on the ACD1K dataset. Camouflaged objects present unique challenges due to their visual similarity with the background, making conventional detection techniques less effective. Our system integrates both static image analysis and real-time live camera detection, augmented with a robust authentication module and user history management. Comprehensive evaluation using metrics such as mAP, IoU, precision, recall, and FPS indicates that our model achieves high detection accuracy (mAP ≈ 85%) and realtime performance (≈20 FPS) even in challenging environments. The proposed methodology not only improves object localization in cluttered scenes but also paves the way for practical applications in surveillance, wildlife monitoring, and security systems [1], [2].
User Authentication: Ensuring only authorized users access the system. YOLOv8 Model: Performing real-time segmentation-based detection. Static Image and Live Camera Processing: Allowing image uploads and real-time detection. User History Management: Storing and analyzing past detections. The main contributions of this research include: 1. Integration of YOLOv8 segmentation with a specialized COD dataset (ACD1K). 2. Dual-mode detection (static images + live camera feeds). 3. User authentication and history tracking for performance monitoring. 4. Performance evaluation using state-of-the-art metrics.
Key Words: Camouflaged Object Detection, YOLOv8, ACD1K, Image Segmentation, Real-Time Detection, Authentication, User History Management
2. COMPARISON WITH EXISTING WORK
1.INTRODUCTION
Several techniques have been explored in camouflaged object detection:
Camouflaged object detection (COD) is an advanced computer vision problem where the target object shares similar visual features with the background, making it difficult to identify. This is particularly relevant in military applications, wildlife observation, and security surveillance.
Method
Traditional object detection models such as Faster RCNN, SSD, and Mask R-CNN have struggled with detecting camouflaged objects due to their reliance on clear object-background separability. However, the emergence of deep learning-based segmentation techniques, particularly the YOLO series, has significantly improved real-time object detection. The YOLOv8 segmentation model, which enhances feature extraction and bounding box regression, is leveraged in our research to detect camouflaged objects more effectively.
© 2025, IRJET
|
Impact Factor value: 8.315
Strengths
Limitations
Faster R-CNN
High accuracy in Computationally standard object expensive, slow detection. inference time.
Mask R-CNN
Not optimized for Instance segmentation camouflaged object with detailed masks. detection.
SSD (Single Faster than Shot Detector) based models. DeepLabV3+
R-CNN- Struggles with small or highly blended objects.
Strong semantic Requires segmentation computational performance. resources.
high
ACDNet (COD- Trained specifically for Lacks real-time specific) camouflaged objects. inference capabilities.
|
ISO 9001:2008 Certified Journal
|
Page 89