Subprojects
The project has few subprojects. Some of them are developed in parallel and others are part of the end of the global project
Last update on February 10th, 2013.
Algorithms
In autonomous vehicle applications each resource, a node, need to exchange observations and data to estimate their pose. Also it is important that each node be able to predicts, not only its own pose, also other nodes ones. Due to bandwidth restrictions, node capacity (hardware and power) and perception needs, this is not a simple task.
Visual perception algorithms
Visual perception and understanding of the environment with video cameras is a research area in continuous growth. The cameras have greatly favored the use of mobile robots in many applications The automotive industry, for example, has invested many efforts to provide some intelligence to their vehicles in order to assist the driver in dangerous situations and thus reduce the risk of accidents. The use of cameras has a key role here, since all traffic signs and other related signals are designed exclusively to be seen by people. However, these systems typically have a high degree of dependence on the structure of the roads and must deal with real ambient conditions such as drastic illumination changes, poor visibility conditions and wide range of environments and infrastructure, among other things. Moreover, the human visual system proves to be robust and flexible enough to allow an experienced driver to successfully drive in extreme conditions, attracting new developments inspired in human vision.
Visual attention and human behaviour
Contact: Marcelo L. Moreyra, marcelo_moreyra_at_ieee.org
It is well studied that visual attention patterns and the associated eye movements reflect cognitive processes. There have been many research works on the role of eye movements in executing everyday visually guided behaviors, where eye trackers measurements were used. Our work is focused on learning about how humans visually inspect an scene while trying to identify a road topology. It is of our interest to get some clues about whether there are specific places in the scene that are strategic to determine the topology of the road, and in which moment the observer pays attention to them. This work also proposes to observe if there is any effect of the saliency of the image in the eye behaviour during the task. With all this information the intention is to develop new algorithms that can interpret non linear roads topologies
The following paper and data is related with this efforts. An experiment was conducted with a group of people who had to observe a set of road photographs and determine for each scene the corresponding topology, e.g. a curve to the left, a T-intersection, a merge or a roundabout, among others. Eye fixations were recorded with an eye-tracker while the participants were solving the assigned task. The evidence obtained here will be a fundamental guide to future efforts in the design of an image processing algorithm for the estimation of nonlinear road topologies. This experiment is of particular interest to research areas including the development of autonomous vehicles perception systems and driver assistance systems.
Papers
Marcelo L. Moreyra, Favio R. Masson, Eduardo. M. Nebot; Visual Attention Patterns for the Perception of Nonlinear Road Topologies in Unstructured Scenes. Submited to ..., Feb 2013
Dataset
If you like to play with the data, click here. If you use them for a paper or research, please refer the owner properly as is indicated in the page.