One Method for Dynamic Reconstruction on Sparse Observation System

One method was proposed in this paper which can update observed data sparsely and reconstruct the scene model densely so that the monocular vision system can observe and reconstruct the target dynamically. On one hand, by combining the classic solutions for observation systems and introducing the coherence-detection and the maintenance-judgment into it, sparse updating and optimization was fulfilled. On the other hand, by introducing adaptive constrain on scattered points into dense reconstruction phase so that the degree of reconstruction-fault tolerance was promoted. Meanwhile, by using a hybrid modeling approach and introducing the gradient-motion model, the interface of the scene was optimized. The experiment result shows that the efficiency of the system can be ensured by separation of reconstruction and observation, the system can basically satisfy the requirement of online analysis on dynamic scene.