Abstract
In the early 90's, image sensors were dominated by CCD technology and CMOS sensors were only developed in research labs. Then a balance within the market shares has been reached thanks to the huge improvement in CMOS integrated circuits fabrication process. This is closely related to the emergence of portable devices like mobile phones, which most often embed photo or video functions. Indeed, such system integration in addition to cost constraints have favored CMOS technology. Nevertheless, video shots with these portable devices, very shaking prone, require a video stabilization which needs to estimate the inter frame global motion of sequences. Therefore, the goal of this work is to add<br />this function to the imagers fabricated by STMicroeletronics.<br />To do so, a new global motion estimation technique is presented in this thesis. This method consists in extracting a global motion model from the local movements perceived in the periphery of images. This has been first validated by an algorithmic approach, before being integrated on silicon. The final architecture of the sensor has a photosensitive area divided into a central area and a peripheral one. The signal processing chain contains a pixel level processing to measure the peripheral local motions. It includes also a post-processing<br />dedicated to the global motion model estimation and to the unwanted motion compensation.