TNO PreScan

The section of Automotive I work in deals with active and passive safety. Think for example of designing vehicle state-estimators for usage with traction control, modeling human body behavior for usage in (dummy) crash test scenarios and developing algorithms for collective adaptive cruise control. Of course, all this has to be tested before it is used in real life. This is done by various simulation software tools. One of them, PreScan, has been the work of me for about four years now.

In short it is a simulation application which is used to verify the validity of various sensor-algorithms for use in vehicles. For example, at what sensor inputs should a parking-assist algorithm start to give signals to the driver? What is the influence of reflectivity due to, for example, rain on a white-line recognition algorithm?

The main assignment was to design and implement various new features for the existing application. For the last three and a half years I have been working on this in a multidisciplinary team at the TNO office in Helmond. During this time the application has been successfully deployed to various customers. A recent question from one of these customers sparked the initiation of a new sub-project for the PreScan application. The question was to have very realistic images as input for their camera sensor system.

We already used a visualization engine based on OpenGL to show the PreScan scenario in 3D, but the output of this system wasn’t realistic enough by far. It couldn’t, for example, use more than 3 light sources. Besides, reflection of glass and/or mirrors was poorly handled. It quickly became clear that, in order to produce a photo-realistic output, something like a ray-tracer had to be used. After some investigation we chose the POV-Ray ray-tracer as our engine and our challenge then became to feed it with realistic models based on the PreScan scenario.