This project focuses on implementing state estimation to track a human subject within a dynamic point cloud environment. Developed for the Autonomous Systems module and inspired by research into 4D Visual Grounding, the system employs a Standard Kalman Filter (KF) under a Constant Velocity motion model. A critical component of the measurement model is the Sonata (Point Transformer v3) neural network. In this pipeline, the Sonata model is responsible for segmenting the human subject from the larger point cloud scene, allowing the system to extract the centroid $(x, y, z)$ as the measurement vector.
🚶 3D Human Trajectory Tracking in Point Clouds using a Standard Kalman Filter
Implementation of Kalman Filter for tracking humans using 3D point cloud data.