In the field of driving assistance applications five major research topics have been identified: sensorial perception (sensorial signal processing), modeling and estimation of the road, modeling and estimation of the obstacles, obstacle classification, and world model representation.
Each of these research topics have their specific problems that we propose to solve. In the field of sensorial perception one aims at modeling the noise, the acquisition errors and the radiometric response to lighting conditions, adapting the sensor to environmental conditions, calibration and fusion of information into a unified description.
In the field of road modeling and estimation one aims at creating relevant models, having a high level of generality but which allow efficient real-time parameter estimation, designing feature extraction techniques by sensorial signal processing, and probabilistic-based robust estimation.
In the field of generic object (structures above ground that can be hit by the ego vehicle) modeling and estimation, one aims at the precise modeling of position, size, shape, orientation and dynamics of these objects, extraction and grouping relevant features by taking into account the sensor’s limitations in precision, consistency, lighting response, and at designing probabilistic estimators for object model parameters estimation.
In the field of classification, one follows the design of a probability based meta classifier, which uses multiple features extracted from the unified sensorial model and from the geometrical, hierarchical and dynamic models of the objects, and combines information from several primary classifiers in order to infer by probabilistic methods the class of the object from several available classes. The relevant classes will be defined along with the relevant features for each class, and the primary classifiers will be established along with their specialized learning algorithms.
As the complexity of the traffic situations handled by the driving assistance systems increases, along with the systems’ degree of autonomy, a unified theoretical handling of the world model becomes absolutely necessary. The world model has to integrate in situ and a priori information about the driving environment in a fashion convenient to driving assistance systems.
From the theoretical and practical study of the field, and from our own experience which was materialized into original results, we can define three major directions where we propose to bring original theoretical contributions: designing new mathematical models for the road, objects and classes, designing algorithms for relevant feature extraction by monocular, color and stereo sensorial information processing, and designing robust probability-based inference methods for road, object and class estimation.
The main objectives of this project are:
- O1. Comprehensive theoretical and practical study of the approached fields.
- O2. Designing original mathematical models for sensorial system performance, road and obstacle geometry and dynamics, object classes and world representation.
- O3. Designing original algorithmic solutions for sensorial information processing and integration, for extraction of relevant features, for road and object parameter estimation, for object classification, and world model representation.
- O4. Design and implementation of a working prototype, testing of algorithms and defining the system’s performance model.
Each of these main objectives has a set of secondary objectives which will organize the research activities. The secondary objectives for O1 are:
- O1.1. Study of the relevant scenarios
- O1.2. Study of the recent approaches in the field
- O1.3. Qualitative and quantitative evaluation of modern modeling techniques in the field
- O1.4. Qualitative and quantitative evaluation of recent algorithms in the field.
Secondary objectives for O2:
- O2.1. Modeling the performance and functionality of the sensorial system.
- O2.2. Designing new models for environment classes and world model representation
- O2.3. Designing probabilistic inference models based on sensorial information.
Secondary objectives for O3:
- O3.1. Defining and development of the unified multimodal sensorial representation.
- O3.2. Design and implementation of estimation and classification algorithms.
- O3.3. Defining the estimator interaction models.
Secondary objectives for O4:
- O4.1. Design and implementation of a working prototype for demonstration and evaluation of the sensorial system from the viewpoint of the driving assistance applications.
- O4.2. Defining the performance models for the algorithms and world representation.
- O4.3. Publishing of results.
The research activity that will follow these objectives will be multidisciplinary. The models that will be created will have solid mathematical foundations, and the algorithm design will start from mathematical solutions. When modeling the dynamics and interactions between objects, road and own vehicle, mechanics and automatics knowledge will be required. From the field of computer science, the knowledge and contributions of this project will involve the disciplines of Image processing, Artificial Intelligence and Shape Recognition Systems.
The project is expected to have a major impact in the field, as it is aimed at development of new concept and solutions that will open new directions of theoretical and applicative research, as well as the development of industrial systems.