3D Object Tracking Method And Applications

3D object tracking is a fundamental problem to a lot of applications in industry au-tomation,virtual reality and augmented reality.We study two related topics,3D rigid body real-time tracking and 3D hand real-time tracking,in this field.Recognition and tracking of panels in assembly process is our major motivation for 3D rigid body real-time tracking.The process has two stages:recognition and tracking.During the recognition stage,reference and target are located in different coordinate systems.For every key point,we extract local geometry features,encode them to local descriptors and transfer the problem of finding the correspondence in 3D space to the problem of finding the nearest neighbor in feature space.During the tracking stage,we take the advantage of the tracking result of previous frame,and align the point of the current frame to the previous reference,and update the orientation and position.We utilize GPU parallel computing to accelerate this process.3D hand real-time tracking is a critical technique in human-computer interaction,which estimates the pose of hand using data captured by sensors.The system can be divided into two components:(1)forward kinematics based 3D hand model,(2)3D hand tracking module.In the first part,we design a 3D hand model based on the kinematics features of human hand.Approximating a human hand using spheres to reduce computing complexity,and converting the task of finding point-surface correspondence to finding point-sphere correspondence.In the second part,by optimizing a carefully design energy function,we fit the model with captured data.Particle swarm optimization and Articulated ICP algorithm are combined to solve the energy function.