Vision and vision sensors play an ever expanding role in pretty much every field of electronics, including security, industrial automation, medical equipment, virtual reality (and augmented and mixed reality), automobiles, and drones – as well as many more. In many of these applications, multiple, often dissimilar, sensors are needed such as, image, radar, IR and time of flight (TOF) etc.
As an example, consider the architecture of an automobile with ADAS features (drones, AR/VR all have similar challenges). The tasks of surround view, parking assistance, traffic sign recognition, emergency breaking, collision avoidance, cross traffic alert, blind spot detection, cruise control and forward vehicle detection are handled by a variety of image sensors, lidar and radars. Application processors typically do not have enough IOs to support this many sensors. A method is needed to aggregate these multiple signals into fewer streams of data.
MIPI CSI-2℠ virtual channels provide the ability to uniquely designate data types and sensor sources within a single CSI-2 video stream. This enables a single video stream to transport video, data and metadata from a variety of sensors. This in turn will minimize the need for multiple cables transmitting information over a long distance, in addition to minimizing the physical IO connections to application processors.
This presentation discusses the architectures and tradeoffs in mission critical vision applications. It also covers the details of combining and tagging multiple data streams, including the Camera Control Interface (CCI) integration.
Tom Watzka is the technical mobile solutions architect at Lattice Semiconductor with more than 20 years of experience developing embedded products, including seven years developing consumer mobile solutions. Currently, Watzka is the marketing product manager for the CrossLink video bridge product line, focused on mobile and mobile influenced markets. He received his bachelor of science degree from the Rochester Institute of Technology and his master's degree from Pennsylvania State University, where he conducted his master’s thesis on FFT algorithms.