The advancement in sensing technology created a need for developing data fusion techniques. In This paper we address the problem of fusing 3D LiDAR data with visual imagery. The purpose of our fusion scheme is the classification of LiDAR data into four different classes. We establish a decision-level fusion scheme to solve this problem using manifolds. Our method proceeds in three steps. First, We used features extracted from each modality to learn a separate manifold. In the second step, we defined a classification confidence level for each class on each manifold using training data. Finally, in order to predict the class of new data point, we predict the class of that point on each manifold separately. Using the classification confidence, We established a decision-level scheme that combines the individual prediction on each manifold into a final prediction. We test our method using two data sets. Results show the effectiveness of our approach for decision-level fusion of multi-modal data.
展开▼
