DynamicVisionCore: A Predictive Object Tracking Framework for Real-Time Robotics Applications

Abstract

DynamicVisionCore is a novel predictive object tracking framework designed for real-time robotics applications. The system integrates YOLOv8 for object detection and DeepSORT for multi-object tracking, ensuring high accuracy and low latency. The framework achieves an impressive MOTA of 72.5% and MOTP of 80.1% on the MOT17 dataset, outperforming traditional methods such as SORT and ByteTrack. Motion prediction is enhanced using a hybrid Kalman Filter and LSTM-based model, reducing RMSE from 9.7 to 6.3 in occlusion scenarios. Additionally, the system ensures robust performance under challenging conditions, achieving 91.3% accuracy in normal lighting and 85.6% in low-light environments. The implementation is optimized for edge computing platforms like Jetson Xavier, where it achieves real-time processing at 9.7ms per frame. The system's robustness is further validated against varying lighting conditions, occlusions, motion blur, and extreme environments. This research demonstrates the efficacy of DynamicVisionCore in real-time robotic vision applications, ensuring reliable object tracking for autonomous systems. Future work will explore <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">a</sup>daptiv<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">e</sup> reinforcement learning strategies and improved sensor fusion techniques to further enhance tracking robustness.