The recent developments in automated driving and driver assistant systems have driven the development of highly sophisticated algorithms for image recognition and data analysis. Many of these algorithms are based on algorithms and design of artificial intelligence (AI). The availability of large databases like vehicle databases, road scenes split into different semantic areas, crash test databases and damage databases originally collected for different targets, make it possible to apply learning algorithms on them and derive different knowledge-based algorithms use-ful for accident reconstruction. This paper presents how these approaches can be used in applications for accident reconstruction. This paper presents the basic principles of neural networks and how self-learning is achieved in these systems in the first part. In the second part practical examples of convolutional neural networks (CNNs) are presented, which are used for semantic segmentation of images and point clouds, image processing (de-blur and superresolution), video analysis, vehicle identification and damage estimation (EES-estimation). Algorithms of artificial intelligence are extremely powerful in situations, where reference datasets are available for the learning phase, but closed formulations of algorithms are very difficult to find doing the job.

(EVU-members can download the full article)