The Pattern Recognition Lab main research topics are
This research topic deals with the optimization of complex vision systems.
Vision system configuratorThe developed configurator is an expert module that improves the ease of use of versatile industrial vision systems. It assists the operator by automatically providing optimal parameter settings for a new configuration of the working environment. It relieves him from the tedious task of system setup. In operation, the configurator learns from a small number of examples and derives a system setting with optimal expected performance. The developed assistant is currently operational in the case of the visual inspection of IC markings. The benefit is an improved and faster system setup. |
Badly configured vision system |
Modern range cameras provide means to measure 3D surfaces easily and automatically, but also set the need for new 3D data interpretation methods.
3D scene analysis3D vision surpasses the limitations of classical video vision in applications like control. surveillance and security. This ends difficulties with shadows and other foreign illumination effects. |
Pedestrian detection by 3D vision |
From real to virtual objectsModeling 3D objects requires the measurement of the object under different views and the subsequent assembly of the views. 1) The measurement is performed by a 3D-scanner that acquires range images 2) Several range images corresponding to different views are needed in order to obtain measurements that cover the complete object 3) During assembly, the various views must be brought together. It requires an exact positioning of the geometry and the merging of the data meshes derived from each range image. The final model describes the full geometry and the color of the object. Applications of 3D modeling are numerous: industrial design, reverse engineering, simulation, 3D photography, building and updating virtual environments, object recognition, ... |
Reverse engineered model of a watch frame |
This research topic aims at improving scene analysis tasks by an efficient design of the early stage of the vision process. Several modes and cues of perception are considered. Also the vision process is composed of a preattentive stage followed by another stage of attentive vision. The interest focuses on visual attention and other related subjects.
Visual attentionAn ongoing activity considers the task of extended visual attention by investigating attention models, which consider scene depth as a perception mode that complements visual perception. The study is oriented towards a new model of multimodal perception as well as its use for solving segmentation and scene understanding tasks. |
Spot of visual attention obtained by a multimodal computer model |
Other ongoing activities concern:
Character recognitionCharacter recognition is best performed on-line when processing is directly part of the sensing device. Several recognition methods are available as candidate for operation in a miniaturized device. Following figure illustrates the recognition by a feed-forward neural network. |
On-line acquired character |
Robot visionBecause the environment of a robot is often unpredictable and changing, there is a strong need to fit out robots with sensing capabilities like vision. We investigate the integration of vision in the frame of the behavioural architecture for the design of mobile robots. |
Mobile robot navigation |
Spray quality inspectionJudging the quality of spray jets is performed routinely by human visual inspection. In this project, we investigated means to perform the task automatically by image analysis. We demonstrated its feasibility and designed a prototype system. In this inspection task, image analysis offers the advantage of an objective, non-destructive and flexible inspection method. |
Spray jet shape analysis |