Return Home

ICT technologies are now beginning to have a major impact in the automotive industry, a major global industry. Increasingly, the basic functions in motor vehicles are becoming computer controlled, and telecommunication is being used to add new functionality to vehicles. A major challenge for automobile makers is to make vehicles safe. Intelligent vehicles could make a significant impact in making our roads safer and reducing the road toll.

Goal

The purpose of the Smart Cars project is to use computer technology, particularly image analysis, to make for a safer driving experience. That goal of the project is not to create an autonomous vehicle, but to provide input to the driver about road conditions, obstacles and potential hazards, letting the driver retain control of the vehicle while benefiting from the input from the computer system. Such systems are generally referred to as Advanced Driver Assistance Systems (ADAS).
 

Research

ADAS requires detailed information about the road scene, the vehicle itself and the driver to analyse the situation properly. We refer to the analysis of the situation as Road Scene Threat Assessment (RSTA). In this project, we are mainly focusing on gathering information from the road scene and the subsequent RSTA.

The research considers

• Detection and tracking of objects in the road scene (cars, pedestrians, animals, signs etc). Particularly how algorithms addressing these problems can be made robust to real-world conditions like fog, rain, shadows, occlusions etc. while at the same time work when both the camera/other sensor and the object is moving. Partially this can be addressed by using additional sensors like radar or cameras in the infrared spectrum, which opens up research issues like how to adequately fuse these.

• Real-time performance with constraints on available computing power. We are targeting the automotive market which means that any system we create must fulfil the criteria of being very cheap, small and not consume vast amounts of energy, yet deliver top performance. By implementing the necessary algorithms in parallel hardware like FPGA’s it is possible to fulfil these criteria. However, it opens up a whole new spectrum of research issues on how to best utilize this hardware. That is, algorithms that have never been considered in traditional CPU’s due to computational complexity may suddenly become viable in parallel hardware, or algorithms that suits traditional CPU’s can not be implemented efficiently in an FPGA.

• Road Scene Threat Assessment. RSTA represents one of the aims of this project, that is, what we do with the information gathered by our sensory systems. The open research issues are how to best analyse the given set of information and extract useful information to pass on to the driver (or perhaps to the vehicle itself if there is an immediate action needed like braking). For information to be useful to the driver, it needs to be given ahead of time of a dangerous situation. Calculating probabilities of potential outcomes in the near future given a set of detected objects is possible, but may lead to an explosion in computational complexity. The question is how to appropriately model a given situation to reduce complexity while at the same time make accurate predictions.

Achievements

• New and improved visual features for efficient detection of objects in cluttered environments have been developed. Apart from many publications in relevant international conferences and journals, our road sign detection technology is currently at such a level that we are able to commercialise it, and we have also demonstrated real-time pedestrian detection from a moving vehicle.

• Design of a computing platform using FPGA’s that serves as a research platform and demonstrator for parallel algorithms. In addition, the corresponding library of computer vision algorithms running on this hardware can be used to implement not only systems to be used on the road, but in other “Smart Camera”-applications.

• Initial road scene threat assessment system validated on real data captured by Volvo (40.000km)

The Smart Cars project is also currently collaborating with CSIRO on far infrared technologies relevant to automatic detection of animals in the road scene.