Structure from motion or stereoscopy are used to obtain 3D from a sequence of still images. However if there is no texture or features in the images, no 3D can be obtained. Featureless environments are difficult to reconstruct in 3D only using cameras. Projected light patterns can be used to measure the shape or an object. However, scattering is the main problem in light based underwater sensors such as light and cameras. Collimated light such as laser minimizes this problems by focusing the light in fewer points.
Publication type: Conferences
Vision based motion estimation algorithms are widely used in ground-based and aerial robotics. Combined with inertial measurement units, they have proven to be a precise and low-cost sensor for velocity and pose estimation. In this paper we show that stereo vision based odometry can be used by autonomous underwater vehicles (AUV) that navigate close to the seabed for velocity and incremental pose estimation in small areas. We present the integration of two different stereo visual odometry algorithms into an AUV and experiments carried out in laboratory and harbour conditions comparing vision based pose estimates with ground truth.
An automatic classifier algorithm has been designed to assess the population of Posidonia oceanica over a set of underwater images at Palma Bay. Laws’ energy filters and statistical descriptors of the Gray Level Co-occurrence Matrix have been use to correctly classify the input image patches in two classes: Posidonia oceanica or not Posidonia oceanica. The input images have been first reprocessed and splitted in three different patch sizes in order to find the best patch size to better classify this seagrass. From all the attributes obtained in these patches, a best subset algorithm has been run to choose the best ones and a decision tree classifier has been trained. The classifier was made by training a Logistic Model Tree from 125 pre-classified images. This classifier was finally tested on 100 new images. The classifier outputs gray level images where black color indicates Posidonia oceanica presence and white no presence. Intermediate values are obtained by overlapping the processed patches, resulting in a smoother final result. This images can be merged in an offline process to obtain density maps of this algae in the sea.
An appearance-based approach for visual mapping and localization is proposed in this paper. On the one hand, a new image similarity measure between images based on number of matchings and their associated distances is introduced. On the other hand, to optimize running times, matchings between the current image and previous visited places are determined using an index based on a set of randomized KD-trees. Further, a discrete Bayes filter is used for predicting loop candidates, taking into account the previous relationships between visual locations. The approach has been validated using image sequences from several environments. Whereas most other approaches use omnidirectional cameras, a single-view configuration has been selected for our experiments.
Robotic systems can achieve real-time visual odometry by extracting a fixed number of invariant keypoints from the current camera frame, matching them against keypoints from a previous frame, and calculating camera motion from matching pairs. If keypoints are selected by response only they can become concentrated in a small image region. This decreases the chance for keypoints to match between images and increases the chance for a degenerate set of matching keypoints.
Here we present and evaluate a simple grid-based method that forces extracted keypoints to follow an even spatial distribution. The benefits of this approach depend on image quality. Real world trials with low quality images show that the method can extend the length of a correctly estimated path by an order of magnitude. In laboratory trials with images of higher quality we observe that the quality of motion estimates can degrade significantly, in particular if the number of extracted keypoints is low. This negative effect can be minimized by using a large number of grid cells.