An Embedded Solution to Visual Mapping for Consumer Drones

Abstract

In this paper, we propose a real-time visual mapping scheme which can be implemented on a low-cost embedded system for consumer-level ratio control (RC) drones. In our work, a 3-dimensional occupancy grid map is obtained based on an estimated trajectory from data fusion of multiple on-board sensors, composed of two downward-facing cameras, two forward-facing cameras, a GPS receiver, a magnetic compass and an inertial measurement unit (IMU) with 3-axis accelerometers and gyroscopes. Taking the advantages of the low-cost FPGA and ARM NEON intrinsics, we run our visual odometry and mapping algorithms at 10Hz on board. Meanwhile, we also present a hierarchical multi-sensor fusion algorithm to provide a robust trajectory for mapping usage. Finally, we verify the feasibility of our approaches and serval potential applications with experimental results in complex indoor/outdoor environments.

Guidance: A visual sensing platform for robotic applications

“…1b shows the quadrotor equipped with Guidance, in which case the players can enjoy GPS-denied hovering and anti-collision features. Technically, Guidance is in fact an upgraded version of Zhou et al's work [31] which is a visual mapping solution based on four cameras and a single processing chip -Altera's SoC FPGA. The block diagram of Guidance's built-in functionalities can be found in Fig.…”

“…Then, there are two threads for image remapping: (1) visual odometer for the localization system and (2) visual mapping for the sense and avoid system. In the first thread, FAST [25,26] feature detector, BRIEF [5] feature descriptor and local binary matching are employed for image feature extraction and association based on the analysis in [31]. The algorithms of motion estimation and multi-sensor fusion will be discussed in detail later.…”

“…The built-in algorithms can be categorized into: visual odometer, visual mapping and multi-sensor fusion. As a matter of fact, the algorithms for pixel-level processing, visual mapping and multi-sensor fusion are introduced in [31]. Here we just want to highlight that the visual odometer used in Guidance is a hybrid version of an inertial-assisted stereo visual odometer discussed in [32] and an inertial-assisted monocular visual odometer which can contribute whenever the stereo system is invalid.…”