Autonomous agricultural systems are a promising solution to bridge the gap between labor shortage for agriculture tasks and the continuing needs for increasing productivity in agriculture. Automated mapping and navigation system will be a cornerstone of most autonomous agricultural system. Accordingly, we propose a ground-level mapping and navigating system based on computer vision technology (Mesh Simultaneous Localization and Mapping algorithm, Mesh-SLAM) and Internet of Things (IoT), to generate a 3D farm map on both the edge side and cloud. The innovation of this system includes three layers as sub-systems that are 1) ground-level robot vehicles’ layer for conducting frames collection only with a monocular camera, 2) edge node layer for image feature data edge computing and communication, and 3) cloud layer for general management and deep computing. High efficiency and speed of mapping stage are enabled by making the robot vehicles directly stream continuous frames to their corresponding edge node. Then each edge node, that coordinate a certain range of robots, applies a new Mesh-SLAM frame by frame, whose core is reconstructing the features map by a mesh-based algorithm with scalable units and reduce the feature data size by a filtering algorithm. Additionally, the cloud-computing allows comprehensive arrangement and heavily deep computing. The system is scalable to larger-scale fields and more complex environment by taking advantage of dynamically distributing the computation power to edges. Our evaluation indicates that: 1) this Mesh-SLAM algorithm outperforms in mapping and localization precision, accuracy, and yield prediction error (resolution at centimeter); and 2) The scalability and flexibility of the IoT architecture make the system modularized, easy adding/removing new functional modules or IoT sensors. We conclude the trade-off between cost and performance widely augments the feasibility and practical implementation of this system in real farms.

Real-time capturing of vehicle motion is the foundation of connected vehicles (CV) and safe driving. This study develops a novel vehicle motion detection system (VMDS) that detects lane-change, turning, acceleration, and deceleration using mobile sensors, that is, global positioning system (GPS) and inertial ones in real-time. To capture a large amount of real-time vehicle state data from multiple sensors, we develop a dynamic time warping based algorithm combined with principal component analysis (PCA). Further, the designed algorithm is trained and evaluated on both urban roads and highway using an Android platform. The aim of the algorithm is to alert adjacent drivers, especially distracted drivers, of potential crash risks. Our evaluation results based on driving traces, covering over 4000 miles, conclude that VMDS is able to detect lane-change and turning with an average precision over 76% and speed, acceleration, and brake with an average precision over 91% under the given testing data dataset 1 and 4. Finally, the alerting tests are conducted with a simulator vehicle, estimating the effect of alerting back or front vehicle the surrounding vehicles’ motion. Nearly two seconds are gained for drivers to make a safe operation. As is expected, with the help of VMDS, distracted driving decreases and driving safety improves.

Indoor localization and navigation have a great potential of application, especially in large indoor spaces where people tend to get lost. The indoor localization problem is the fundamental of an indoor navigation system. Existing research and commercial efforts have leveraged wireless-based approaches to locate users in indoor environments. However, the predominant wireless-based approaches, such as WiFi and Bluetooth, are still not satisfactory, either not supporting commodity devices, or being vulnerable to environmental changes. These issues make them hard to deploy and maintain. In this paper, we present Vivid, a mobile device-friendly indoor localization and navigation system that leverages visual cues as the cornerstone of localization. By leveraging the computation power at the extreme internet edges, Vivid to a large extent overcomes the difficulties brought by resource-intensive image processing tasks. We propose a grid-based algorithm that transforms the feature map into a grid, with which finding the path between two positions can be easily obtained. We also leverage deep learning techniques to assist in automatic map maintenance to adapt to the visual changes and make the system more robust. With edge computing, user privacy is preserved since the visual data is mainly processed locally and detected dynamic objects are removed immediately without saving to databases. The evaluation results show that: i) our system easily outperforms the existing solutions on COTS devices in localization accuracy, yielding decimeter-level error; ii) our choice of the system architecture is scalable and optimal among the available ones; iii) the automatic map maintenance mechanism effectively ameliorates the localization robustness of the system.

Precision agriculture aims to use minimal inputs to generate maximal yields by managing the plant and its environment at a discrete instead of a field level. This new farming methodology requires localized field data including topological terrain attributes, which influence irrigation, field moisture, nutrient runoff, soil compaction, and traction and stability for traversing agriculture machines. Existing research studies have used different sensors, such as distance sensors and cameras, to collect topological information, which may be constrained by energy cost, performance, price, etc. This study proposed a low-cost method to perform farmland topological analytics using sensor implementation and data processing. Inertial measurement unit sensors, which are widely used in automated vehicle study, and a camera are set up on a robot vehicle. Then experiments are conducted under indoor simulated environments that include five common topographies that would be encountered on farms, combined with validation experiments in a real-world field. A data fusion approach was developed and implemented to track robot vehicle movements, monitor the surrounding environment, and finally recognize the topography type in real time. The resulting method was able to clearly recognize topography changes. This low-cost and easy-mount method will be able to augment and calibrate existing mapping algorithms with multidimensional information. Practically, it can also achieve immediate improvement for the operation and path planning of large agricultural machines.