Computer Vision News - April 2021

248 Computer Vision Application Alejandro Galindo is VP of Research & Development at Iris Automation. Iris Automation builds detect-and-avoid systems for industrial and commercial drones. He speaks to us about their innovative work which is bringing together the conventional worlds of computer vision and geometry with the latest advances in artificial intelligence and deep learning. To integrate safely into civilian airspace, drones must be able to detect obstacles and get out of their way as early as possible to avoid collision. This means being able to see other aircraft that may be a kilometer or more in the distance. Detect-and- avoid technology powers unmanned aircraft with the ability to understand what is in their surroundings and to take the safest course of action to keep the airspace safe. Practically, this is complex for drones because of their size, weight and power . In the past, there have been attempts to develop solutions using RADAR, but for drones that is too power-hungry and not aerodynamic. LIDAR is also unsuitable because it only provides short-range accuracy. Some have tried to use sound, but that won’t detect certain aircraft, such as balloons or gliders. Recent advances in science and technology have allowed this work to progress at a pace. Products such as NVIDIA Jetson devices have been a huge enabler for the robotics and autonomous vehicles industry, while deep learning has led to giant leaps forward in computer vision. Alejandro Galindo asserts there is a way to merge more traditional techniques with the cutting edge. His team have developed a solution which uses visual spectrum imaging with cameras and fuses geometry and conventional computer vision techniques with the state of the art in deep learning . High-resolution imagery is acquired in real time, at least 10 frames per second, and processed on an embedded computer Alejandro Galindo