The proposed navigation system is built around the hypothesis that it can function on a smartphone or smartglasses device (like Google Glass or Epson Moverio BT300) using only their built-in sensors (video camera, internal IMU: Inertial Measurement Unit, and GPS: Global Positioning System).
The considered industrial application is an interactive pedestrian guiding system mounted on smartglasses for the visually impaired.
THE PEDESTRIAN TIME TO BEAT PORTABLE
As a matter of fact, many vision-based embedded system prototypes using multiple sensors and cameras exist for the visually impaired and blind involving heavy portable GPU equipment and sensors. Less popular than these types of systems discussed in recent the literature, the pedestrian egocentric navigation system has very similar problems. Real-time computer vision-based scene understanding tasks can be very complex problems especially for navigational systems like autonomous driving and drone flying as the onboard camera is moving asynchronously to its environment and the system needs to adapt to new paths, sceneries, and obstacles. This paper focuses on the Time-to-Collision network’s ability to adapt to new sceneries with different types of obstacles with supervised learning. It consists of two modules: a static data extractor made of a convolutional neural network to predict the obstacle position and distance and a dynamic data extractor that stacks the obstacle data from multiple frames and predicts the Time-to-Collision with a simple fully connected neural network. We propose a novel light weighted and time-efficient vision-based solution to predict Time-to-Collision from a monocular video camera embedded in a smartglasses device as a module of a navigation system for visually impaired pedestrians. As difficult vision-based tasks like object detection and monocular depth estimation are making their way in real-time applications and as more light weighted solutions for autonomous vehicles navigation systems are emerging, obstacle detection and collision prediction are two very challenging tasks for small embedded devices like drones.
0 kommentar(er)
