VisLab ’ s Experience on Autonomous Driving in Extreme Environments

The laboratory of Artificial Vision and Intelligent System of Parma University, that recently has started an academy spin-off called VisLab, has been involved in basic and applied research developing machine vision algorithms and intelligent system for the automotive field for more than one decade. The experience gathered through these years covers a wide range of competencies, either under the algorithmic and technological point of view. VisLab developed, tuned and tested several image processing algorithms, such as: obstacle detection, obstacle avoidance, path detection, terrain estimation and classification; often these results have been achieved by fusion between vision and active sensor, for a better 3D reconstruction of the surrounding environment. On the technological side VisLab can rely on a long experience in using several types of vision sensors (FIR, NIR, visible) as well as on laser sensors (both mono and multi plane), multi echo sensors, together with the support of odometric and inertial information. One of the most distinctive feature that characterize VisLab, that explains its continuous presence in project with manufactures of vehicles of various types, is the very specific experience in developing and setting up out-of-laboratory unmanned prototypes. The application of vision system on board of autonomous vehicles not only requires to fully dominate the cutting edge vision technologies, but also to have a deep experience in dealing with unknown, and often hostile, environments and corresponding challenges: calibration, illumination variation, noise, temperature, installation requirements, etc . Beside its expertise, the key to VisLab’s quick application prototyping is the proprietary software GOLD, that has been developed in the last 10 years and that constitutes the basis of each application developed. GOLD is, at the same time, a powerful development framework for image processing application and a software capable of real-time data acquisition, synchronization, logging, and—ultimately—data processing and visualization, for improved researcher efficiency.