A real time approach to the recognition of human activity based on the interaction with objects in domestic settings is presented. It is based on capturing partial images where the activity takes place using a color camera, and processing the images to recognize the present objects and their location changes. For object description and recognition, a histogram on a normalized chromaticity space has been selected. The interaction with the objects is classified as four types of possible actions regarding change in location (add, remove, move or unchanged). Activities are defined as receipts, where objects plays the role of ingredients, tools or substitutes. Sensed objects and actions are then used to analyze in real time the probability of the human activity that is being performed at the moment in a continuous activity sequence. Tests are carried on a automated kitchen scenario where proactive assistance wants to be implemented taking into account the early recognition user activity.

An assistive rehabilitation strategy for a lower-limb wearable robot is proposed and evaluated. The control strategy monitors the human-orthosis interaction torques and modifies the orthosis operation mode depending on its evolution with respect to a normal gait pattern. The control algorithm relies on the adaptation of the joints stiffness in function of these interaction torques and to the deviation from the desired trajectory. A walking pattern, an average of recorded gaits, is used as reference input. The human-orthosis interaction torques are used to define the time instant when robot assistance is needed and its degree. The objective of this work is to demonstrate the feasibility of ensuring a dynamic stability by means of an efficient real-time stiffness adaptation for multiple joints and simultaneously maintaining their synchronization. The algorithm has been tested with five healthy subjects showing its efficient behavior in maintaining the equilibrium while walking in presence of external forces. The work is performed as a preliminary study to assist patients suffering from Spinal cord injury and Stroke.

A new method for obtaining a three-dimensional volumetric reconstruction from a set of views improving the classical Shape from Silhouette method (SFS) is presented. SFS approaches can be easily accelerated through hardware and software techniques but they are very sensible to errors arising during calibration and segmentation processes so they present difficulties when dealing with real images. This paper proposes a new algorithm which uses the information about pixel segmentation uncertainty contained in each view in order to get a reliable 3D reconstruction of the scene. Aggregation of the projected uncertainties permits to classify scene's voxels by means of a decision rule but also makes it possible to create a three-dimensional confidence map of the scene. As a consequence, the regions where more information is needed can be foreseen. Sample reconstructions from real image sets are presented and evaluated.

Perspective projection presents objects as they are naturally seen by the eye. However, this type of mapping strongly distorts their geometric properties as angles, which are not preserved under perspective transformations. In this work, this angular variation serves to model the visual effect of perspective projection. Thus, knowing that the angular distortion depends on the point of view of the observer, it is demonstrated that it is possible to determine the pose of an object as a consequence of its perspective distortion. It is a computational approach to direct perception in which spatial information of a scene is calculated directly from the optic array. Experimental results show the robustness provided by the use of angles and establishes this 3D measurement technique as an emulation of a visual perception process.

A diversity of action-perception applications relies on 2D information to perform 3D tasks. Minimally invasive surgery (MIS) is one of these applications. It requires a high degree of sensory-motor skills to overcome the disengagement between action and perception caused by the physical separation of the surgeon with the operative site. The integration of body movements with visual information serves to assist the surgeon providing a sense of position. Our purpose in this paper is to present the exterior orientation as a tool in assisted interventions locating the instruments with respect to the surgeon. An enhanced perception is obtained by augmenting the 2D information imposing position and orientation data through a human-machine interface. Applying motion analysis in a sequence of images and having knowledge of the 3D transformations implemented to the instrument, we show it is possible to estimate its orientation with only two different rotations and also its position in the case its length information is supplied.

Robotic surgery has experienced a rapid progress in a diversity of interventions, especially in minimally invasive surgery (MIS). In this type of interventions the patient body is accessed by inserting especial instruments through small incisions. As these instruments are manipulated by robotic assistants, passive wrists are incorporated for free compliance with the port of entry. With this wrist configuration it is only possible to locate accurately an instrument tip if its port of entry or fulcrum point is known. A vision oriented method is proposed in this paper to estimate the position of this external point of incision. It is based on motion analysis in a sequence of images where geometric properties of projected features and 3D transformations of the instrument are related. In this work we show that it is possible to accurately estimate the position and orientation of this 3D point, as well as the orientation of the instrument with respect to the laparoscopic camera applying determined movements through the robotic assistant.

Like oral communication, gestures are a natural way to carry out human machine interface. In the early days of robotic systems, human gesture was used to control robot movements by means of a master-slave structure. In spite of the use if robot programming languages, manual control is the most reliable way to carry out complex tasks in unstructured environments. In these situations, a non-contact, passive and remote system can be helpful to control a teleoperated robot by means of human gestures. In this paper, a vision system able to detect, locate and track the head and hands of a human body is presented. The system uses several calibrated cameras placed around the operator scenario to locate the body parts of a person in 3D. The system combines different computer vision techniques to increase the reliability of the body parts detection: image movement detection, user skin colour segmentation and stereo. The data provided by these modules are focused looking for coherence according to the human body dimensions. With the scheme proposed it is possible to obtain a low-cost real-time system for human computer interfacing based in a natural way of communication (gestures). Civil area such as big robots in shipyards, mines, public works or cranes is some possible applications.

The aim of the work is to determine the best landmark for its use in tasks such as precise positioning of AGVs or mobile robots in front of loading/unloading areas. The work developed aims at determining the precision that can be achieved with different pattern landmarks, of use in positioning systems, based on computer vision. Different computer vision algorithms have been tested from the images acquired from different camera configurations and bending angles with respect to the vertical axis. As a result of the evaluation of these errors, we extract the recommendation of the most favourable bicolour pattern to be used, as well as the bounding errors obtained from the segmentation and recognition methods most frequently used.

Object recognition systems need image segmentation processes that relate image regions to world objects. These methods present often three problems: the generation of a large number of small regions, undersegmentation (different objects are associated to the same image region) and oversegmentation (a scene object is segmented in various regions). In order to overcome these problems, we propose an image segmentation method that combines depth information and object surface properties obtained from a pair of stereo images. The system work under the standard assumption that 3D objects have planar faces and regular shapes. First a region growing segmentation process is applied to both images generating two labeled images. Then, depth information of the region frontiers is obtained by matching the labeled segments from left and right image rows. The stereo matching problem is solved by finding a path through a 2D search plane whose axes are the left and right segmented lines. Original image regions are then merged based on their size, surface information and frontiers depth information. In this way, image regions are associated to surfaces that are contiguous in the 3D space and they present a common property (such as gray level, color or texture).

Huge and expensive computation resources are usually required to perform graph labelling at high speed. This fact restricts an extensive use of this methodology in industrial applications such as visual inspection. A new systolic architecture is presented which computes structural distances between cliques of different graphs based on a modified incremental Levenshtein distance algorithm. The distances obtained are used as a support function for graph labelling using probabilistic relaxation techniques. The proposed architecture computes the distances between k input cliques of an input graph and one reference clique of a reference graph. It does not limit the number of cliques nor cliques complexity of the input graph, so any input graph can be labelled. A low cost solution has been implemented based on FPGAs.