Object Detection Method Using Image and Number of Objects on Image as Label

Aim:

           To develop an object detection model using YOLOv8 to address the limitations of existing methods and improve detection accuracy, robustness, and efficiency. The aim is to design a system that reduces the dependency on extensive labelling while ensuring adaptability to unseen environments. The model will utilize YOLOv8’s capabilities to process data efficiently and deliver high-performance results for diverse applications in object detection.

Abstract:

         The proposed system focuses on object detection leveraging YOLOv8’s advanced capabilities. Unlike traditional methods requiring extensive labelled data, the proposed model aims for high accuracy with optimized labelling requirements and adaptability to diverse environments. The system includes dataset collection, model building, and a prediction mechanism, offering enhancements in detection performance, reduced computational requirements, and versatility in real-world scenarios. YOLOv8’s cutting-edge features ensure robust results even in complex scenarios, making it a suitable choice for real-time applications.

Introduction:

        Object detection is a cornerstone in computer vision, powering applications like surveillance, autonomous driving, and robotics. While deep learning-based approaches such as CNNs and transformers have revolutionized this field, challenges like reliance on bounding box annotations and adaptability to unseen environments persist. Traditional object detection systems often require significant computational resources and heavily annotated datasets, which limit their scalability. The proposed system employs YOLOv8, a state-of-the-art model, to overcome these challenges and deliver superior detection results. This model combines efficiency, accuracy, and adaptability to meet modern application requirements.

Proposed System:

           The proposed system integrates YOLOv8’s advanced features for streamlined object detection. Key features include:

• Use of images with minimal labelling to reduce preparation costs.
• Enhanced adaptability to unseen data, ensuring robustness across various scenarios.
• Efficient prediction and classification pipeline, enabling real-time performance.
• Incorporation of a lightweight architecture to minimize computational requirements.

The system addresses existing challenges by focusing on performance optimization and usability, making it suitable for a wide range of applications.