A novel edge-enabled SLAM solution using projected depth image information
Jian-qiang Li,Yi-fan Zhang,Zhuang-zhuang Chen,Jia Wang,Min Fang,Cheng-wen Luo,Huihui Wang
DOI: https://doi.org/10.1007/s00521-019-04156-2
2019-05-08
Neural Computing and Applications
Abstract:Environmental mapping is the key step for mobile robots to perform tasks independently and perfectly. In recent years, visual SLAM, laser-based SLAM and simultaneous localization and mapping (SLAM) have aroused the interest of many people. Unfortunately, those technologies are not widely used, limited by the computational complexity, data processing and very low and predictable latency. This paper had mainly completed the following work and edge-enabled computing-based edge computing (Shi and Dustdar in Computer 49(5):78–81, 2016) is used as a solution to accelerate calculation. First of all, this research design works with inertial unit mobile robotic navigation systems, and all sensors are connected in edge layers in the framework of edge computing and explore the accelerometer, electronic compass, and gyroscope data. The accelerometer data are integrated using the Kalman filter data fusion algorithm to filter the random drift error caused by the gyroscope and the electronic compass. The state of the machine is determined by calculation of the corresponding attitude angle and position information. Second, a low-cost distance sensor is used to detect the depth and upload to the other fog node for computation. Next, the 3D point coordinate information is projected onto the two-dimensional coordinate extraction feature point to establish the feature map. Third, the extended Kalman filter SLAM is used to achieve simultaneous positioning and mapping. Finally, the method is validated in the experiment, proving that the method is feasible. The main improvement in this article is as follows: First, the multi-sensor data fusion algorithm is used to reduce the positioning error. Second, we use low-cost distance sensors to measure the depth of the model environment and reduce the cost. Third, we would take advantage of translating the three-dimensional depth information into a flat two-dimensional projection information to reduce the calculation of load and computing time. Fourth, our computation is distributed in different layers and focuses on edge-enabled platform to decrease the latency and redundancy.
computer science, artificial intelligence