Loading...
Please wait, while we are loading the content...
Similar Documents
Panoramic and active camera for object mapping color and depth-based superpixels for semantic image segmentation.
| Content Provider | CiteSeerX |
|---|---|
| Author | Jebari, Islem Filliat, David |
| Abstract | We present an approach to multi-objective semantic segmentation based on depth and color. Our goal is to build a semantic map containing high-level information, namely furniture and objects. This approach was developed for the Panoramic and Active Camera for Object Mapping (PACOM) [1] project in order to participate in a French exploration and mapping contest called CAROTTE [2]. Our problem is defined as follows: given an input image, we assign a label to each one of the pixels; the labels are associated to high-level concepts that give a semantic interpretation to the scene; adjacent components with the same label constitute the ”semantic segments” and are associated to the real world objects indicated by the label. We present an evaluation of different segmentation algorithms for this task. Color-Depth Superpixels for Semantic Segmentation The implemented algorithm uses a Markov Random Fields (MRF) to assign labels to the superpixels. The process is divided in two main phases: training and testing. During training, the MRF is trained using a set of labeled, color and depth images. During testing, the MRF is used to assign labels to new images. RGB images Depth images Fig. 1. Kinect database Acquisition Prediction: given a set of observations to infer the most probable assignations for the latent variables. |
| File Format | |
| Access Restriction | Open |
| Subject Keyword | Active Camera Semantic Image Segmentation Object Mapping Color Depth-based Superpixels Markov Random Field High-level Concept Implemented Algorithm Real World Object Adjacent Component Multi-objective Semantic Segmentation Depth Image Semantic Interpretation Mapping Contest Color-depth Superpixels Semantic Segment New Image Probable Assignation Different Segmentation Algorithm Database Acquisition Prediction Label Constitute High-level Information Input Image Latent Variable Main Phase Semantic Segmentation Object Mapping French Exploration Semantic Map Depth Image Fig |
| Content Type | Text |
