Navigation of an AUV for investigation of underwater structures

There is a great demand for autonomous underwater vehicles (AUVs) to investigate artificial underwater structures such as piles and caissons in harbours, and risers and jackets of deep-sea oilfields. This paper proposes an autonomous investigation method of underwater structures using AUVs that is implemented by initially detecting the target objects, localizing them, then approaching them by taking video images while closely tracing their shape. A laser ranging system and a navigation method based on the relative position with respect to the target objects are introduced to realize this behaviour.     

Proximal object and hazard detection for autonomous underwater vehicle with optical fibre sensors

This paper describes short range and tactile optical fibre sensors for marine applications. The sensors are designed for obstacle avoidance on unmanned underwater vehicles (UUVs) operating in confined spaces, but have other possible applications. The fibre sensors augment the sensory abilities derived from ultrasonic and other sensors employed for marine proximity measurement. Of particular interest is proximity detection in the “near” (less than 1 m) and tactile areas. The paper describes the basic principle of operation and alternative sensor configurations. Results are given based on laboratory tests and deployment on a mini autonomous submersible in a test pool.     

Mission design for a group of autonomous guided vehicles

In this paper, a generic approach for the integration of vehicle routing and scheduling and motion planning for a group of autonomous guided vehicles (AGVs) is proposed. The AGVs are requested to serve all the work stations in a two-dimensional environment while taking into account kinematics constraints and the environment’s geometry during their motion. The problem objective is the simultaneous determination of time-optimum and collision-free paths for all AGVs. The proposed method is investigated and discussed through a number of simulated experiments using a variety of environments and different initial conditions.     

Accurate 3D-vision-based obstacle detection for an autonomous train

In this paper we present a 3D-vision based obstacle detection system for an autonomously operating train in open terrain environments. The system produces dense depth data in real-time from a stereo camera system with a baseline of 1.4 m to fulfill accuracy requirements for reliable obstacle detection 80 m ahead. On an existing high speed stereo engine, several modifications have been applied to significantly improve the overall performance of the system. Hierarchical stereo matching and slanted correlation masks increased the quality of the depth data in a way that the obstacle detection rate increased from 89.4% to 97.75% while the false positive detection rate could be kept as low as 0.25%. The evaluation results have been obtained from extensive real-world test data. An additional stereo matching speed-up of factor 2.15 was achieved and the overall latency of obstacle detection is considerably faster than 300 ms.     

An efficient system for combined route traversal and collision avoidance

Here we consider the problem of a robot that must follow a previously designated path outdoors. While the nominal path, a series of closely spaced via points, is provided with an assurance that it will lead to the destination, we can’t be guaranteed that it will be obstacle free. We present an efficient system capable of both following the path as well as being perceptive and agile enough to avoid obstacles in its way. We present a system that detects obstacles using laser ranging, as well as a layered system that continuously tracks the path, avoiding obstacles and replanning the route when necessary. The distinction of this system is that compared to the state of the art, it is minimal in sensing and computation while achieving high speeds. In this paper, we present an algorithm that is based on models of obstacle avoidance by humans and show variations of the model to deal with practical considerations. We show how the parameters of this model are automatically learned from observation of human operation and discuss limitations of the model. We then show how these models can be extended by adding online route planning and a formulation that allows for operation at varying speeds. We present experimental results from an autonomous vehicle that has operated several hundred kilometers to validate the methodology.

Numerical comparison of steering geometries for robotic vehicles by modeling positioning error

This paper describes an analytical method for modeling the positioning error of a robotic vehicle and examines how the metric of this error can be used to compare the geometries of various steering configuration. Positioning error can be caused by many factors stemming from the robot’s hardware and software configurations and the interaction between the robot and its environment. A slip motion model that captures the effects of key factors that contribute to positioning error is presented. Robot kinematic models with and without slippage are reformulated and used to perform an in-depth assessment and characterization of positioning error. The method is applied to three characteristic advance and steering configurations: Ackermann, articulated, and explicitly steered. This analysis serves as a quantitative evaluation of the properties of the steering geometries for path tracking under identical slippage conditions. The method can also be used as a tool for comparing robot configurations to make trade-off decisions early in the design process, as it allows for derivation of predicted performance values of alternative steering geometries.

Frequency response method for terrain classification in autonomous ground vehicles

Many autonomous ground vehicle (AGV) missions, such as those related to agricultural applications, search and rescue, or reconnaissance and surveillance, require the vehicle to operate in difficult outdoor terrains such as sand, mud, or snow. To ensure the safety and performance of AGVs on these terrains, a terrain-dependent driving and control system can be implemented. A key first step in implementing this system is autonomous terrain classification. It has recently been shown that the magnitude of the spatial frequency response of the terrain is an effective terrain signature. Furthermore, since the spatial frequency response is mapped by an AGV’s vibration transfer function to the frequency response of the vibration measurements, the magnitude of the latter frequency responses also serve as a terrain signature. Hence, this paper focuses on terrain classification using vibration measurements. Classification is performed using a probabilistic neural network, which can be implemented online at relatively high computational speeds. The algorithm is applied experimentally to both an ATRV-Jr and an eXperimental Unmanned Vehicle (XUV) at multiple speeds. The experimental results show the efficacy of the proposed approach.

Analysis of natural images processing for the extraction of agricultural elements

This work presents several developed computer-vision-based methods for the estimation of percentages of weed, crop and soil present in an image showing a region of interest of the crop field. The visual detection of weed, crop and soil is an arduous task due to physical similarities between weeds and crop and to the natural and therefore complex environments (with non-controlled illumination) encountered. The image processing was divided in three different stages at which each different agricultural element is extracted: (1) segmentation of vegetation against non-vegetation (soil), (2) crop row elimination (crop) and (3) weed extraction (weed). For each stage, different and interchangeable methods are proposed, each one using a series of input parameters which value can be changed for further refining the processing. A genetic algorithm was then used to find the best value of parameters and method combination for different sets of images. The whole system was tested on several images from different years and fields, resulting in an average correlation coefficient with real data (bio-mass) of 84%, with up to 96% correlation using the best methods on winter cereal images and of up to 84% on maize images. Moreover, the method’s low computational complexity leads to the possibility, as future work, of adapting them to real-time processing.     

Basic navigation,guidance and control of an Unmanned Surface Vehicle

This paper discusses the navigation, guidance and control (NGC) system of an Unmanned Surface Vehicle (USV) through extended at sea trials carried out with the prototype autonomous catamaran Charlie. In particular, experiments demonstrate the effectiveness, both for precision and power consumption, of extended Kalman filter and simple PID guidance and control laws to perform basic control tasks such as auto-heading, auto-speed and straight line following with a USV equipped only with GPS and compass.

Feature correspondence and motion recovery in vehicle planar navigation

This paper describes a strategy to feature point correspondence and motion recovery in vehicle navigation. A transformation of the image plane is proposed that keeps the motion of the vehicle on a plane parallel to the transformed image plane. This permits to define linear tracking filters to estimate the real-world positions of the features, and allows us to select the matches that accomplish the rigidity of the scene by a Hough transform. Candidate correspondences are selected by similarity, taking into account the smoothness of motion. Further processing brings out the final matching. The methods have been tested in a real application.     

Source seeking with non-holonomic unicycle without position measurement and with tuning of forward velocity

We consider the problem of seeking the source of a scalar signal using an autonomous vehicle modeled as the non-holonomic unicycle and equipped with a sensor of that scalar signal but not possessing the capability to sense either the position of the source nor its own position. We assume that the signal field is the strongest at the source and decays away from it. The functional form of the field is not available to our vehicle. We employ extremum seeking to estimate the gradient of the field in real time and steer the vehicle towards the point where the gradient is zero (the maximum of the field, i.e., the location of the source). We employ periodic forward–backward movement of the unicycle (implementable with mobile robots and some underwater vehicles but not with aircraft), where the forward velocity has a tunable bias term, which is appropriately combined with extremum seeking to produce a net effect of “drifting” towards the source. In addition to simulation results we present a local convergence proof via averaging, which exhibits a delicate periodic structure with two sinusoids of different frequencies—one related to the angular velocity of the unicycle and the other related to the probing frequency of extremum seeking.     

The ANFIS approach applied to AUV autopilot design

This paper describes the application of neurofuzzy techniques in the design of autopilots for controlling the yaw dynamics of an autonomous underwater vehicle. Autopilots are designed using an Adaptive-Network-based Fuzzy Inference System (ANFIS), a chemotaxis tuning methodology and a fixed fuzzy rule-based approach. To describe the yaw dynamic characteristics of an autonomous underwater vehicle, a realistic simulation model is employed. Results are presented which demonstrate the superiority of the ANFIS approach. It is concluded that the approach offers a viable alternative method for designing such autopilots.     

Experimental study on autonomous mobile robot acquiring optimal action to avoid moving obstacles

The principal aim of this study was to show how an autonomous mobile robot can acquire the optimal action to avoid moving multiobstacles through interaction with the real world. In this paper, we propose a new architecture using hierarchical fuzzy rules, a fuzzy evaluation system, and learning automata. By using our proposed method, the robot autonomously acquires finely tuned behavior which allows it to move to its goal and avoid moving obstacles by using the steering and velocity control inputs simultaneously. We also show experimental results which confirm the feasibility of our method.     

自然场景下的文本定位

首先阐述了文本定位的基本流程,然后列举了现有的主要文本定位方法,分析了基于区域、纹理、边缘、角点的文本定位方法和机器学习的文本定位方法的优缺点,详细说明了文本区域验证和文本块区域合并的方法,最后总结了各种文本定位方法。     

基于生物视觉特征的目标轮廓提取算法

自然场景中的目标轮廓提取是计算机视觉中的一个重要研究问题。其难点在于场景中大量的纹理边缘严重地干扰了轮廓提取的完整性。近年来,一些研究工作将生物视觉特征引入图像边缘轮廓提取,取得了一定的效果。其中通过引入视觉外区抑制特征可以在提取物体轮廓边缘的同时抑制一定量的纹理边缘,从而得到轮廓边缘集合。然而在整合轮廓边缘时,传统模型仅仅采用求交并集的简单合并方法,使得强响应的细小纹理残留。基于此,提出了一种改进的基于生物视觉特征的自然场景目标轮廓提取算法。首先采用多水平抑制方法得到候选轮廓边缘集合。接着将一种基于生物视觉特征的边缘组合方法用于将候选边缘整合成为一个完整的目标轮廓。与传统的外区抑制算法相比,基于视觉特征的轮廓提取算法提高了自然场景中目标轮廓提取的准确性和完整性。     

改进YOLOv8的无人机航拍图像目标检测算法

针对无人机航拍图像存在多个小目标聚集、目标尺度变化大的问题;提出一种改进YOLOv8的目标检测算法TS-YOLO（tiny and scale-YOLO）。在主干部分去除冗余的特征提取层;设计了一种高效特征提取模块（efficient feature extraction module;EFEM）;避免小目标特征消失在冗余信息中。在颈部设计了一种双重跨尺度加权特征融合方法（dual cross-scale weighted feature-fusion;DCWF）;融合多尺度信息的同时抑制噪声干扰;提升特征表达能力。通过构建一种参数共享检测头（parameter-shared detection header;PSDH）;使回归和分类任务实现参数共享;保证检测精度的同时有效降低了模型的参数量。所提模型在VisDrone-2019数据集上的精度（P）和召回率（R）分别达到54.0%、42.5%;相比于原始YOLOv8s模型;mAP50提高了5.0个百分点;达到44.5%;且参数量减少了55.8%;仅有4.94×106;在DOTAv1.0遥感数据集上;mAP50达到71.9%;仍具有较好的泛化能力。