Intelligent Systems LabUniversidad Carlos III de Madrid

Computer Vision Systems

3D Vision

    Obtaining 3D measurements can be performed in different ways: using a camera and a laser, using two or more cameras and more recently, using 3D cameras also provide color information.

    Our group LSI has experience using 3D technology, it has been performed real time large objects, calibration and synchronization of multiple cameras and the use of RGB-D cameras for modeling environments.

Vision Applied to Robots

    Computer vision has led to a more flexible use of robots, allowing them to perceive and interpret the environment. The works carried out have focused on the identification of artificial and natural markings, which help in the tasks of navigation and location robot.

    Cheaper and reduced hardware size allows growing perception that the algorithms can be executed in real time on the robots.

Object Recognition using Geometric Model

    In some cases the shape of objects can be represented by the union of geometric shapes such as lines, rectangles, ellipses, etc. To determine how well a particular model fits the image to the edges of it are obtained, the distance to them, their symmetry, color or gray level areas that include, establishing a formulation which measures the degree of adequacy.


IR Images

    Far infrared (FIR) can analyze scenes with adverse lighting conditions by capturing the heat from objects. However, images usually have low contrast, low resolution and noise than can be achieved with good lighting conditions.

    Our algorithms are able to recognize pedestrians and determine their direction of travel analyzing their gait.


Machine Learning Algorithms

    Machine learning can learn from and make predictions on data. Training a computer to automatically learn to know what are the characteristics that define an object and how should then be sought in the picture to find regardless of their position, scale and rotation.

    This approach proved very useful for the detection of pedestrians with cameras visible and infrared spectrum. Detection of people is a challenge given the wide variety of appearances and ways that may appear in pictures.


Features Analysis: Color & Texture

    Color is a feature that can help you recognize objects. It presents difficulties demand greater computational cost and vary greatly to changes in lighting. Algorithms have been made for the detection and classification of traffic signs using color and considering the frequent changes due to weather conditions and the diversity of colors that have traffic signals.

    Thanks to the development of computers and the increase in computing power and can perform quality checks based on objects that have texture. That is no longer necessary that the intensity has to be uniform, so, in products such as wood, cloth, tiles, defects can be detected or be categorized depending on the spatial variation of their properties.


Surveillance Systems

    Systems have been made for the detection and monitoring of persons variable lighting environments. Thus vision system transmits the images only when someone enters an unauthorized zone.For a human operator is a tedious task, with this system, operator can monitor multiple cameras at the same time.

  • Multi-Object Tracking

Multi-Object Tracking by a visual-based identification of the objects. Unlike systems using the learning of an appearance model for every tracked object, the proposed data association method is based on only one similarity model capable of discriminating whether a detected person in the current frame corresponds, or not, to one of the tracked ones. This method allows an online tracking of any target, selected by a user over a live sequence, without requiring previous knowledge of it.

  • Person Re-Identification

Three-dimensional representation to compare person images, which is based on the similarity, independently measured for the head, upper body, and legs from two images. Three deep Siamese neural networks have been implemented to automatically find salient features for each body part.

  • Deep Learning

A novel normalized double margin-based contrastive loss function for the training of Siamese networks, which not only improves the robustness of the learned features against the intra-class variations and inter-class ambiguities but also reduce the training time.

Quality Control

    It has developed a system of quality control for metal surfaces, applied to the automotive sector, in order to detect imperfections in the body before being painted.