Unmanned Systems Technology 027 l Hummingbird XRP l Gimbals l UAVs insight l AUVSI report part 2 l O’Neill Power Systems NorEaster l Kratos Defense ATMA l Performance Monitoring l Kongsberg Maritime Sounder

12 August/September 2019 | Unmanned Systems Technology Nvidia has developed a six-camera system that tracks detected objects as they appear in consecutive camera images by assigning them unique ID numbers (writes Nick Flaherty). Camera object tracking is an essential component of the surround camera vision pipeline of an autonomous vehicle, as it plays a critical role in robust distance-to-object and object velocity estimations. This can also mitigate missed and false-positive object detections and prevent them from propagating into the planning and control functions of a self-driving car. These functions are used to make decisions about stopping and accelerating.  The system uses six cameras to avoid blind spots around the car. The software in it tracks objects in all six camera images, and associates their locations in image space with unique ID numbers as well as time-to-collision (TTC) estimates. The surround camera object tracker predicts the object scale change, translation and TTC using specific points on the objects, and allows a latency of around 90 ms. This enables the car to respond properly to sudden stopping conditions, even in high-speed driving. It also addressed the problem of ghost tracks of objects moving out of the scene, which are the main cause of false- positive braking. The tracker removes ghost tracks within 30-150 ms without causing object track misses. However, running object detection on all six camera feeds for each image frame can increase the system’s overall latency. Depending on the application, developers can choose to reduce object detection frequency and run the object detector once every three frames, for example. That would free up resources for other computations. The system successfully tracked another car at a distance of 150 m with a 100 mph relative velocity differential, with the object detection running once every three frames. The surround object tracking system has been successfully tested for more than 20,000 miles with a car in autonomous mode. The tests were performed in various seasons, routes, times of day, illumination conditions, highways and urban roads. No disengagements owing to object tracking failures were observed or reported. Nvidia’s DriveWork software development kit contains the object tracking module, which can run on a CPU or GPU to provide flexibility in the overall resource scheduling for the software pipeline. The GPU implementation is seven times faster than the CPU implementation, but the software is flexible enough to have the object tracker on a CPU for one camera and on the GPU for the other five. Vision systems To each object its own ID Object tracking and TTC computation diagram of Nvidia’s camera system Researchers at the Daegu Gyeongbuk Institute of Science and Technology in South Korea have developed a radar system that can detect miniature UAVs at a distance of over 3 km (writes Nick Flaherty). The system, developed by researcher Daegun Oh and his team, operates at between 12 and 18 GHz, and uses a combination of an active electronically scanned array with a type of neural net called a generative adversarial network (GAN). The GAN has two neural nets that compete to produce the best results to pick up UAVs weighing as little as 500 g. Radar can spot mini-UAVs Detection