CHIPS: Chirping Pixels for Self-Powered Visual Sensing

Project AbstractComputer vision has permeated many aspects of our everyday lives. The camera has served as the key source of informa,tion for computer vision. The data that it provides, the image, remains the fundamental measurement of the visual world. This is in,great part because the camera was initially developed to produce images for human consumption. In the context of AI, the consumer of, visual data is a machine. In this realm, one has the opportunity to rethink the imaging paradigm and come up with new sensing metho,ds that deviate substantially from the traditional camera. Our proposed work is a radical departure from traditional imaging. In the, coming decade, we see imaging as becoming a key sensing modality for developing smart environments. In this context, the camera is,often not the most suitable measurement device. First, the camera captures visual information from a single viewpoint. While the ima,ge it captures is rich in details, it often includes more information than needed to accomplish a task. Second, the electronics of t,he camera requires significant power to record images. Third, given the large number of pixels in an image, significant power is als,o needed to transmit the image to a remote location. For all these reasons, cameras are forced to be externally powered and tethered,. Our goal in this work is ambitious. It is to redesign the pixel so that it can measure the intensity of light falling on it withou,t needing any external power. Furthermore, it is able to transmit information related to its intensity measurement wirelessly to a r,emote receiver. Our pixel design is compelling for applications that require a small number of light measurements. The pixels can be, strategically positioned in any given environment without any cables connecting them to power sources or other devices. We describe, initial experiments and lay out a research plan that includes the use of our self-powered pixels for several vision applications.