Object Detection and Classification Framework for Analysis of Video Data Acquired from Indian Roads.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Autonomous Systems

Background:

• Object detection and classification are vital for autonomous vehicle (AV) navigation.
• Existing AV algorithms face challenges with unique Indian road conditions like diverse traffic and weather.
• There is a need for specialized, efficient algorithms for AVs in India.

Purpose of the Study:

• To develop a robust and efficient object detection and classification model for Indian roads.
• To adapt deep learning techniques for real-time AV applications in challenging environments.
• To improve the safety and reliability of AVs operating in India.

Main Methods:

• Utilized the YOLOv8 deep learning model for its lightweight and scalable architecture.
• Conducted experimental evaluations using real-life videos from diverse Indian road scenarios.
• Assessed performance across 35 distinct object classes, considering factors like low lighting and occlusions.

Main Results:

• Achieved a precision of 0.65 for multi-class object detection.
• Demonstrated an average real-time accuracy exceeding 70% across various conditions.
• Reached a peak accuracy of 95% under optimal conditions, outperforming existing methods.

Conclusions:

• The proposed YOLOv8 approach offers a superior balance between model complexity and performance for Indian roads.
• The model is well-suited for deployment in AVs due to its high accuracy and minimal computational requirements.
• This research contributes to advancing AV technology for complex, real-world driving environments in India.