Event based visual codification in automotive environments

In this paper an Address Event Representation (AER) based system for hands driver activity detection in a close-real automotive environment is presented. AER is a biologically inspired technique based on the codification of the visual information in events instead of frames. This approach could be highly appropriate in the case of the transmission of high density information like images in a restrictive automotive-compliant systems, but any application have been presented in this field. This work presents the first complete AER system developed to a real application in an automotive environment. The main goal is to evaluate the performance of this promising technology for the visual information codification and transmission in a cabin track simulator. The presented system is biologically plausible, with a low computational complexity and works with a standard low-cost camera. Two system architecture approaches are presented: synaptic and neural. In the first case AER is applied only for visual codification, and in the second one AER is apply for codification and processing tasks. Different pixel activity detection and AER event generation methods have been deeply evaluated. Results obtained are compared with a classical computer vision system based on JPEG transmission. It has been proved that for the scan AER event generation method the performance over JPEG is improved. As our work results show, AER technology could be an effective alternative in real automotive conditions.