Stereo-vision system design using camera-link for low-voltage automotive CMOS image sensors

Abstract

Machine vision has come of age. As costs drop, it is gradually moving beyond the laboratory and industrial automation applications and is entering our everyday lives offering us the unquestionable benefits of higher security, safety and automation. One of the first domains touted for quick adoption of these systems is the automotive market which is perennially pressured to improve its safety record. SENSATION was an FP6 Project that fostered the creation of sensors including vision, with the aim of improving safety in automotive applications and beyond by monitoring and managing human alertness. Stereovision is one of the techniques proposed by SENSATION for tracking a driver’s head and posture with a third camera for measuring eye-blink rates and saccades. Special hardware is required to meet the high performance and synchronisation criteria between the left and right video streams. A truly integrated stereovision camera is required. This project sets out to fill this need by specifying, designing, developing, implementing, testing and finally deploying such a system using image sensors specifically developed by ATMEL for this purpose. Current techniques used for stereovision synchronisation fall short of what is required for high speed operation which calls for new alternative techniques to be developed and validated. Performance and accuracy are the major concerns but so are reliability, flexibility, power consumption, weight and cost. Any solution must sacrifice none of these and must strike an equitable balance. In this dissertation, the entire engineering process that lead to the creation of an innovative automotive-grade electronic vision product is explored. A streamlined, low-cost stereovision hardware platform that adheres to industry standards and is acceptable by automotive standards is what this project sets out to achieve.