Search

Evaluating the number of berries per bunch is a vital step of grape yield estimation in viticulture but is a labour intensive task for traditional manual measurement. Therefore, this paper develops a novel smartphone application for counting berries automatically from a single image. The application, called 3DBunch, acquires images from the camera or the album on a smartphone, and then estimates the number of berries by a reconstructed 3D bunch model based on the proposed image analysis techniques that are embedded in the developed iOS app. It also presents features of visualising the statistics of the reconstructed bunch, which including of the distribution of detected berry size in pixels and the total number of berries. It also has the capability of presenting sampling related information, which includes the person who conducted the sampling, the location of samples, the dates of sampling, the variety, farm, vineyards etc. The application was evaluated both on a simulator on a commercial computer and an iPad mini 4. By analysing 291 bunch images from two varieties the app achieved an accuracy of 91% regarding berry counting per bunch. Additionally, the computational time consumed to process 100 images on iPad mini 4 was studied and returned an average of 7.51 seconds per image. Obtaining these results with only a smartphone and a small backing board for capturing a photo with a single bunch, 3DBunch provides an efficient way for farmers to count berries in-vivo and it is available in the iOS App Store.

There is presently no solution to the problem of an autonomous bulldozer pushing mounds of material to desired goal locations in the presence of obstacles whilst obeying the kinematic constraints of the bulldozer. Past work has solved some aspects of this problem, but not all. This research presents the first complete, practical solution to the problem. It works by creating a fixed RRT in advance, and then during operation connecting pushing poses into this RRT using Bezier curves. The RRT algorithm leverages a novel data structure for performing nearest neighbour comparisons for Ackermann-steering vehicles; termed the Distmetree. The resulting pushing states are searched using greedy heuristic search to find a solution and the final path is smoothed with cubic Bezier curves. The mode of operation chosen for best performance also constructs bidirectional RRTs to reach difficult to access pushing poses. The final mode of the algorithm was tested in simulation and proven to be able to solve a wide variety of maps in a few minutes while obeying bulldozer kinematic constraints. The algorithm, whilst not optimal, is complete which is the more desirable property in industry, and the solutions it produces are both feasible and reasonable.

Smartphones have several advantages over specialist monitoring systems including ubiquity, price, and ease of implementing updates. Thermal imaging can be used to assess plant water status and allow more informed irrigation decisions; unfortunately, this technique has not been widely adopted due to the high cost of equipment and the lack of a system to provide analysis and results in real-time. Several inexpensive thermal cameras that connect to smartphones have recently been released and one of these (FLIR One) was evaluated as part of a system to assess grapevine water status. Irrigation treatments were established on Cabernet Sauvignon and Chardonnay vines in an arid region. Thermal images were taken from the shaded side of the grapevine canopy and software was developed to automatically determine the temperature of the canopy and artificial reference leaves. The temperature readings and metrological inputs were used to calculate five indices of water status including the Crop Water Stress Index (CWSI) and the stomatal conductance index. The best performing was the CWSI, which does not require input from a weather station. Over 30 days of assessment, and a range of irrigation levels, measurements collected with the thermal camera were correlated with stem water potential (R2 = 0.61) and stomatal conductance (R2 = 0.74). Windy conditions appeared to be the major cause of variation between CWSI and stomatal conductance. Inexpensive thermal cameras have the potential to be an easy and accessible tool for the assessment of plant water status and to make better irrigation decisions.

Imaging beyond RGB bands has the ability to provide solutions for non-destructive detection of objects or scene properties. Methods often used in research include hyperspectral cameras and spectrometers; the former is expensive and the latter can only provide point measurements. Multispectral imaging can provide low-cost solutions in instances where RGB does not suffice, however, evaluation of filters used in multispectral sensors or the use of off-the-shelf filters for user specific applications have not been explicitly analysed. This paper proposes a novel method for using spectrometer data by firstly limiting the search space to existing off-the-shelf filters, modelling those filters, and then applying the desired model for classification or regression to more appropriately model multispectral imaging performance. The results indicate that point measurements produced by spectrometers may not correspond with the response achieved when using off-the-shelf filters; particularly in the instance where the spectral profile of the measured object varies significantly within the transmission regions of the filter. The method presented focuses on the practical estimation of filter performance when selecting filters.

The task of bulldozing has many aspects that need more research before they can be fully automated. This paper presents a new method for calculating a path for vehicles tasked with pushing multiple objects to multiple goal locations. Unlike previous approaches which either focus on a single object, apply both push and pull operations, or operate in discrete domain, this solution works in a continuous space whilst handling multiple objects in a push-only scenario and doesn’t use computationally heavy methods to find a potential path planning solution. To achieve the results, two graph structures, one for the active vehicle and one for the passive pushable objects, is used. These graphs are then combined with an A⁎ Search Algorithm to find a correct path for the bulldozer to take. The algorithm is tested on a Microban problem set which results in a 96% success rate.

Digital image processing is commonly used in plant health and growth analysis, aiming to improve research efficiency and repeatability. One focus is analysing the morphology of stomata, with the aim to better understand the regulation of gas exchange, its link to photosynthesis and water use and how they are influenced by climatic conditions. Despite the key role played by these cells, their microscopic analysis is largely manual, requiring intricate sample collection, laborious microscope application and the manual operation of a graphical user interface to identify and measure stomata. This research proposes a simple, end-to-end solution which enables automatic analysis of stomata by introducing key changes to imaging techniques, stomata detection as well as stomatal pore area calculation. An optimal procedure was developed for sample collection and imaging by investigating the suitability of using an automatic microscope slide scanner to image nail polish imprints. The use of the slide scanner allows the rapid collection of high-quality images from entire samples with minimal manual effort. A convolutional neural network was used to automatically detect stomata in the input image, achieving average precision, recall and F-score values of 0.79, 0.85, and 0.82 across four plant species. A novel binary segmentation and stomatal cross section analysis method is developed to estimate the pore boundary and calculate the associated area. The pore estimation algorithm correctly identifies stomata pores 73.72% of the time. Ultimately, this research presents a fast and simplified method of stomatal assay generation requiring minimal human intervention, enhancing the speed of acquiring plant health information.

Background Stomata analysis using microscope imagery provides important insight into plant physiology, health and the surrounding environmental conditions. Plant scientists are now able to conduct automated high-throughput analysis of stomata in microscope data, however, existing detection methods are sensitive to the appearance of stomata in the training images, thereby limiting general applicability. In addition, existing methods only generate bounding-boxes around detected stomata, which require users to implement additional image processing steps to study stomata morphology. In this paper, we develop a fully automated, robust stomata detection algorithm which can also identify individual stomata boundaries regardless of the plant species, sample collection method, imaging technique and magnification level. Results The proposed solution consists of three stages. First, the input image is pre-processed to remove any colour space biases occurring from different sample collection and imaging techniques. Then, a Mask R-CNN is applied to estimate individual stomata boundaries. The feature pyramid network embedded in the Mask R-CNN is utilised to identify stomata at different scales. Finally, a statistical filter is implemented at the Mask R-CNN output to reduce the number of false positive generated by the network. The algorithm was tested using 16 datasets from 12 sources, containing over 60,000 stomata. For the first time in this domain, the proposed solution was tested against 7 microscope datasets never seen by the algorithm to show the generalisability of the solution. Results indicated that the proposed approach can detect stomata with a precision, recall, and F-score of 95.10%, 83.34%, and 88.61%, respectively. A separate test conducted by comparing estimated stomata boundary values with manually measured data showed that the proposed method has an IoU score of 0.70; a 7% improvement over the bounding-box approach. Conclusions The proposed method shows robust performance across multiple microscope image datasets of different quality and scale. This generalised stomata detection algorithm allows plant scientists to conduct stomata analysis whilst eliminating the need to re-label and re-train for each new dataset. The open-source code shared with this project can be directly deployed in Google Colab or any other Tensorflow environment.

Background and Aims Rapid and non‐destructive techniques for measuring grape composition have the potential to improve harvest decisions and to allow the batching of fruit into parcels of similar composition to improve processing efficiency. If such techniques are to be based on the non‐destructive assessment of visible berries, then relationships between the composition of visible berries and that of the whole bunch will need to be described. Methods and Results Total soluble solids, pH, TA and the concentration of anthocyanin, phenolic substances and tannin of visible berries and associated entire bunches of Shiraz and Chardonnay were measured over three seasons. Good correlations were obtained between the composition of visible berries and that of the entire bunch. Most relationships were not 1:1 but were consistent and stable across seasons. Conclusions The accuracy and stability across the seasons of relationships between the composition of visible berries and that of the entire bunch for both Chardonnay and Shiraz were evaluated. Significance of the Study The approaches used in this study to predict bunch composition from visible berry composition could aid the development of non‐destructive techniques based on sensing visible berries to assess grape composition during ripening.

Automated flower counting systems have recently been developed to process images of grapevine inflorescences, which assist in the critical tasks of determining potential yields early in the season and measurement of fruit-set ratios without arduous manual counting. In this paper, we introduce a robust flower estimation system comprised of an improved flower candidate detection algorithm, flower classification and finally flower estimation using calibration models. These elements of the system have been tested in eight aspects across 533 images with associated manual counts to determine the overall accuracy and how it is affected by experimental conditions. The proposed algorithm for flower candidate detection and classification is superior to all existing methods in terms of accuracy and robustness when compared with images where visible flowers are manually identified. For flower estimation, an accuracy of 84.3% against actual manual counts was achieved both in-vivo and ex-vivo and found to be robust across the 12 datasets used for validation. A single-variable linear model trained on 13 images outperformed other estimation models and had a suitable balance between accuracy and manual counting effort. Although accurate flower counting is dependent on the stage of inflorescence development, we found that once they reach approximately EL16 this dependency decreases and the same estimation model can be used within a range of about two EL stages. A global model can be developed across multiple cultivars if they have inflorescences with a similar size and structure.

Smartphones have several advantages over specialist systems for extending crop monitoring technology to small to medium scale horticultural operators, including ubiquity, price, user familiarity and the ease of implementing updates. They also contain sufficient computing power that analysis and support software can be contained within the phone. This work evaluated a range of smartphone based tools for measuring vine water status, leading to the development of the most promising tool into a smart phone application that can be easily used by vineyard managers. Potential systems evaluated included: 1) an infrared camera that is integrated into or connected directly to the smartphone and uses established techniques for the analysis of thermal imagery to assess water status; 2) a portable near infra-red spectrophotometer that interfaces with the phone and measures reflectance across wavelengths for the calculation of water status indices; 3) a 3D camera that is integrated into or connected to the phone via WiFi and can use image analysis to assess the shape or orientation of the leaves; 4) a microscope attached to the smartphone camera or as a separate portable unit that can be used to measure stomatal number and aperture and then calculate stomatal conductance. A trial site with a range of irrigation deficit treatments applied to 'Chardonnay' and 'Cabernet Sauvignon' grapevines was established in the Riverland of South Australia. Water status measurements from the smartphone based sensors described above were benchmarked against conventional methods including mid-day stem water potential and stomatal conductance. The thermal infrared camera system was selected as the most accurate and robust option for development into an app, which will be released to selected viticulturists for beta testing in 2017.

Background Stomatal behavior in grapevines has been identified as a good indicator of the water stress level and overall health of the plant. Microscope images are often used to analyze stomatal behavior in plants. However, most of the current approaches involve manual measurement of stomatal features. The main aim of this research is to develop a fully automated stomata detection and pore measurement method for grapevines, taking microscope images as the input. The proposed approach, which employs machine learning and image processing techniques, can outperform available manual and semi-automatic methods used to identify and estimate stomatal morphological features. Results First, a cascade object detection learning algorithm is developed to correctly identify multiple stomata in a large microscopic image. Once the regions of interest which contain stomata are identified and extracted, a combination of image processing techniques are applied to estimate the pore dimensions of the stomata. The stomata detection approach was compared with an existing fully automated template matching technique and a semi-automatic maximum stable extremal regions approach, with the proposed method clearly surpassing the performance of the existing techniques with a precision of 91.68% and an F1-score of 0.85. Next, the morphological features of the detected stomata were measured. Contrary to existing approaches, the proposed image segmentation and skeletonization method allows us to estimate the pore dimensions even in cases where the stomatal pore boundary is only partially visible in the microscope image. A test conducted using 1267 images of stomata showed that the segmentation and skeletonization approach was able to correctly identify the stoma opening 86.27% of the time. Further comparisons made with manually traced stoma openings indicated that the proposed method is able to estimate stomata morphological features with accuracies of 89.03% for area, 94.06% for major axis length, 93.31% for minor axis length and 99.43% for eccentricity. Conclusions The proposed fully automated solution for stomata detection and measurement is able to produce results far superior to existing automatic and semi-automatic methods. This method not only produces a low number of false positives in the stomata detection stage, it can also accurately estimate the pore dimensions of partially incomplete stomata images. In addition, it can process thousands of stomata in minutes, eliminating the need for researchers to manually measure stomata, thereby accelerating the process of analysing plant health.

A vision system for automated yield estimation and variation mapping is proposed.The proposed method produces F1-score 0.90 in average over four experimental blocks.The developed shoot detection does not require manual labeling to build a classifier.The developed system only requires low-cost off-the-shelf image collection equipment.The best EL stage for imaging shoots is around EL stage 9 regarding yield estimation. Counting grapevine shoots early in the growing season is critical for adjusting management practices but is challenging to automate due to a range of environmental factors.This paper proposes a completely automatic system for grapevine yield estimation, comprised of robust shoot detection and yield estimation based on shoot counts produced from videos. Experiments were conducted on four vine blocks across two cultivars and trellis systems over two seasons. A novel shoot detection framework is presented, including image processing, feature extraction, unsupervised feature selection and unsupervised learning as a final classification step. Then a procedure for converting shoot counts from videos to yield estimates is introduced.The shoot detection framework accuracy was calculated to be 86.83% with an F1-score of 0.90 across the four experimental blocks. This was shown to be robust in a range of lighting conditions in a commercial vineyard. The absolute predicted yield estimation error of the system when applied to four blocks over two consecutive years ranged from 1.18% to 36.02% when the videos were filmed around E-L stage 9.The developed system has an advantage over traditional PCD mapping techniques in that yield variation maps can be obtained earlier in the season, thereby allowing farmers to adjust their management practices for improved outputs. The unsupervised feature selection algorithm combined with unsupervised learning removed the requirement for any prior training or labeling, greatly enhancing the applicability of the overall framework and allows full automation of shoot mapping on a large scale in vineyards.

Women wishing to conceive are largely unaware of fertility symptoms at the time of ovulation. This study investigated the effectiveness of fertility-awareness in achieving pregnancy, particularly fertile mucus pattern, in the context of infertility. The 384 eligible participants were taken from consecutive women desiring pregnancy who attended 17 Australian Billings Ovulation Method

Estimation of phenology distribution in horticultural crops is very important as it governs the timing of chemical thinning in order to produce good quality fruit. This paper presents a novel phenology distribution estimation method named DeepPhenology for apple flowers based on CNNs using RGB images, which is able to efficiently map the flower distribution on an image-level, row-level, and block-level. The image classification model VGG-16 was directly trained with relative phenology distributions calculated from manual counts of flowers in the field and acquired imagery. The proposed method removes the need to label images, which overcomes difficulties in distinguishing overlapping flower clusters or identifying hidden flower clusters when using 2D imagery. DeepPhenology was tested on both daytime and night-time images captured using an RGB camera mounted on a ground vehicle in both Gala and Pink Lady varieties in an Australian orchard. An average Kullback-Leibler (KL) divergence value of 0.23 over all validation sets and an average KL value of 0.27 over all test sets was achieved. Further evaluation has been done by comparing the proposed model with YOLOv5 and shown to outperform this state-of-the-art object detection model for this task. By combining relative phenology distributions from a single image to a row-level or block-level distribution, we are able to give farmers a precise and high-level overview of block performance to form the basis for decisions on chemical thinning applications.

Maintaining metadata records for scientific imaging is challenging where the link between the metadata and the image is labour intensive to create and can easily be broken. We propose a method for using QR codes in images of samples to embed the metadata in an open and robust manner, so that it can be readily extracted on demand. By using a novel pipeline for generating QR codes, displaying them in images, reading the QR codes in the images and extracting the metadata for later action such as renaming the image file, a streamlined process for metadata management is introduced. This method was simulated using a range of image types and QR code parameters to identify the limits of various parameter combinations, providing practical insight into code design and usability. The pipeline was also tested with hundreds of images in both laboratory and field situations and proved to be extremely efficient and robust. This method offers potential for anyone taking images of samples who needs to guarantee the existence and correctness of metadata without relying on an external association mechanism.

This paper presents a method for generating a spatial map of a particular plant or environmental property of a vineyard block based on low cost camera technology and existing vineyard vehicles. Such properties can range from leaf area, per vine bunch count or bare-wire detection. The paper provides a low cost ground vehicle based solution that does not rely on live GPS position recording. Rather, the relative estimated motion between video frames is used to localize each sensor reading within the bounds of each row. Row end locations are derived from post-processed GPS recorded locations of the perimeter of a block with an aerial photograph. This paper uses the proportion of leaf colored pixels in a video frame as a token example of measuring the relative growth of vines during the shoots stage.

Non-productive canopy detection in a viticultural block is a key factor in reducing the drain on infrastructure and improving management practices. However, current methods are significant in cost, biased, and do not provide information on location of non-productive canopy. This paper proposes both a proximal and remote sensing method for assisting in decision support and yield estimation from available technologies. The proximal method utilizes two different measures of green pixel thresholding in video frames, with results providing a useful relative measure of productivity across a vineyard block at the phenological stage of shoots. The remote sensing method utilizes local thresholding and Self-Organizing-Maps on aerial imagery to identify missing vines and total non-productive canopy on a block level. Results indicate the success of this semi-supervised method in providing a useful measure of non-productive canopy at the phenological stage of veraison; laying the groundwork for improved methods in this area. These methods provide practical outputs that lay the foundations for improving management decisions in an automatic and low-cost manner at different times in the season.

Apple trees commonly require the removal of excessive flowers by thinning to produce high quality fruit. Machine vision has recently been applied to detect the flower density as the first step in this process. Existing work relying on color thresholding is sensitive to imaging conditions and the most recent published work using deep learning in this context has proven to be exceptionally slow to process. This paper presents an apple flower segmentation method on a pixel level based on a Fully Convolutional Network (FCN) together with a process of generating a map that can be used for a variable rate chemical sprayer. Despite the challenging conditions of an uncontrolled environment, our apple flower detector was able to generate a F1 score at pixel-level up to 85.6%, which is a relatively high accuracy in terms of pixel-level segmentation. Our method has been tested on both daytime and night-time datasets, which strongly validates the ability of our apple flower detector to work under different conditions. The resulting detections are georeferenced and merged into a density map in the format necessary for application by a variable rate chemical sprayer. Finally, this flower density mapping system will benefit farmers by visualising the whole crop and extracting useful information to support their decision making for chemical thinning.

Counting the number of berries per bunch is a key component of many yield estimation processes but is exceptionally tedious for farmers to complete. Recent work into image processing in viticulture has produced methods for berry counting, however these require some degree of manual intervention or need calibration to manual counts for different bunch architectures. Therefore, this paper introduces a fast and robust calibration-free algorithm for berry counting for winegrapes to aid yield estimation. The algorithm was tested on 529 images collected in the field at multiple vineyards at different maturity stages and achieved an accuracy of approximately 89% per bunch. As it would mostly likely be used to obtain an average value across a block, the low bias of this method resulted in an average accuracy of 99% and was shown to be robust from pea-sized to harvest and between both red and green bunches. Taking only 0.1 to 1 s per image to process and requiring only a smartphone and small backing board to capture, the algorithm can readily be applied to images which are captured in the field by farmers. This allowed bunch weights to be estimated to within 92% accuracy and assisted larger scale yield estimation processes to achieve accuracies of between 3% and 16%. The robustness of the method lays the foundation for fast fruit-set ratio determination and more detailed bunch architecture studies in vivo on a large scale.

Precise yield estimation in vineyards using image processing techniques has only been demonstrated conceptually on a small scale. Expanding this scale requires significant computational power where, by necessity, only small parts of the images of vines contain useful features. This paper introduces an image processing algorithm combining colour and texture information and the use of a support vector machine, to accelerate fruit detection by isolating and counting bunches in images. Experiments carried out on two varieties of red grapes (Shiraz and Cabernet Sauvignon) demonstrate an accuracy of 88.0% and recall of 91.6%. This method is also shown to remove the restriction on the field of view and background which plagued existing methods and is a first step towards precise and reliable yield estimation on a large scale.

Kidnapping occurs when a robot is unaware that it has not correctly ascertained its position, potentially causing severe map deformation and reducing the robot's functionality. This paper presents metric-based techniques for real-time kidnap detection, utilising either linear or SVM classifiers to identify all kidnapping events during the autonomous operation of a mobile robot. In contrast, existing techniques either solve specific cases of kidnapping, such as elevator motion, without addressing the general case or remove dependence on local pose estimation entirely, an inefficient and computationally expensive approach. Three metrics that measured the quality of a pose estimate were evaluated and a joint classifier was constructed by combining the most discriminative quality metric with a fourth metric that measured the discrepancy between two independent pose estimates. A multi-class Support Vector Machine classifier was also trained using all four metrics and produced better classification results than the simpler joint classifier, at the cost of requiring a larger training dataset. While metrics specific to 3D point clouds were used, the approach can be generalised to other forms of data, including visual, provided that two independent ways of estimating pose are available. Two methods for kidnap detection using local pose estimation techniques are proposed.At least two independent ways of estimating relative pose are required.Metrics assessing the quality of a pose estimate are developed and evaluated.For applications with limited training data, a joint classifier performs well.If a large training dataset is available, an SVM classifier is more accurate.

A fast image processing method is proposed for detecting stomata and measuring stomatal aperture size in individual images. The accuracy of aperture measurements is 97%. A prototype mobile application is developed to assist field measurements.

The problem of remote-controlling a mobile robot through the Internet with its associated bandwidth constraints is addressed in this paper. Our solution combines a novel communication and processing module with a unique sensor layout and a flexible control architecture to achieve a range of capabilities from traditional teleoperation to point-and-click autonomy. Careful management of the available bandwidth enables a demonstration of these capabilities between nodes with 20,000 km separation while also providing real-time three-dimensional (3D) models of the environment through the Internet. A spatially oriented compression algorithm, integral to efficient bandwidth management, is also presented. Experiments establish the effectiveness of the extended situational awareness in improving the efficiency and accuracy of driving a mobile robot through a cluttered environment over the existing 2D map or video streaming methods. © 2012 Wiley Periodicals, Inc. (See www.youtube.com/UNSWMechatronics and www.robotics.unsw.edu.au for more information.)

The National Lung Screening Trial (NLST) is a randomized multicenter study comparing low-dose helical computed tomography (CT) with chest radiography in the screening of older current and former heavy smokers for early detection of lung cancer, which is the leading cause of cancer-related death in the United States. Five-year survival rates approach 70% with surgical resection of stage IA disease; however, more than 75% of individuals have incurable locally advanced or metastatic disease, the latter having a 5-year survival of less than 5%. It is plausible that treatment should be more effective and the likelihood of death decreased if asymptomatic lung cancer is detected through screening early enough in its preclinical phase. For these reasons, there is intense interest and intuitive appeal in lung cancer screening with low-dose CT. The use of survival as the determinant of screening effectiveness is, however, confounded by the well-described biases of lead time, length, and overdiagnosis. Despite previous attempts, no test has been shown to reduce lung cancer mortality, an endpoint that circumvents screening biases and provides a definitive measure of benefit when assessed in a randomized controlled trial that enables comparison of mortality rates between screened individuals and a control group that does not undergo the screening intervention of interest. The NLST is such a trial. The rationale for and design of the NLST are presented.

This paper presents Micro Aerial Vehicles (MAVs) and their cooperative systems including Unmanned Ground Vehicles (UGVs) and a Base Station (BS), which were primarily designed for the 1st US-Asian Demonstration and Assessment on Micro-Aerial and Unmanned Ground Vehicle Technology (MAV08). The MAVs are of coaxial design, which imparts mechanical stability both outdoor and indoor while obeying a 30 cm size constraint. They have carbon fibre frames for weight reduction allowing microcontrollers and various sensors to be mounted on-board for tele-operated and waypoint control. The UGVs are similarly equipped to perform their own search and tracking mission but also to support the MAVs by relaying data between the MAVs and the BS when they are out of direct range. The BS monitors the vehicles and their environment and navigates them autonomously or with humans in the loop through the developed GUI. The ability of the MAV in flight was demonstrated by showing continuous hovering. The efficacy of the overall system was validated by autonomously controlling two UGVs for cooperative search.

PurposeThe purpose of this paper is to present a localisation system for an indoor rotary‐wing micro aerial vehicle (MAV) that uses three onboard LEDs and base station mounted active vision unit.Design/methodology/approachA pair of blade mounted cyan LEDs and a tail mounted red LED are used as on‐board landmarks. A base station tracks the landmarks and estimates the pose of the MAV in real time by analysing images taken using an active vision unit. In each image, the ellipse formed by the cyan LEDs is used for 5 degree of freedom (DoF) pose estimation with yaw estimation from the red LED providing the 6th DoF.FindingsAbout 1‐3.5 per cent localisation error of the MAV at various ranges, rolls and angular speeds less than 45°/s relative to the base station at known location indicates that the MAV can be accurately localised at 9‐12 Hz in an indoor environment.Research limitations/implicationsLine‐of‐sight between the base station and MAV is necessary while limited accuracy is evident in yaw estimation at long distances. Additional yaw sensors and dynamic zoom are among future work.Practical implicationsProvided an unmanned ground vehicle (UGV) as the base station equipped with its own localisation sensor, the developed system encourages the use of autonomous indoor rotary‐wing MAVs in various robotics applications, such as urban search and rescue.Originality/valueThe most significant contribution of this paper is the innovative LED configuration allowing full 6 DoF pose estimation using three LEDs, one camera and no fixed infrastructure. The active vision unit enables a wide range of observable flight as the ellipse generated by the cyan LEDs is recognisable from almost any direction.

The current availability of increasingly powerful mobile devices is driving a race to integrate such technology with a diversity of new fields. In order to explore one such largely unexplored application, this paper presents an onboard mobile device-based system for autonomous control of a UGV. Contextualised by the requirements of the Australian AGVC competition, a multi-application software suite was developed for Android, implementing a number of open-source software libraries adapted for a control system application. The developed system is capable of full soft-real time autonomous control, capitalising exclusively on the Android device's embedded sensors for data acquisition. Preliminary results indicated robust capability to perceive, map and plan paths around obstacles, and perform adequately accurate localisation in real time. The effectiveness of the system, considering minimum computational power and hardware cost, not only serves as a proof of application to drive research for the next generations of mobile devices, but garnered considerable interest in competition against classical autonomous system setups.

Precise yield estimation using image processing techniques has been demonstrated conceptually on a small scale. Expanding these solutions to larger scale applications requires significant computational power, which need to analyze the entirety of all captured image data. However, many images captured for yield estimation in these processes only contain small areas of useful features for analysis. This paper introduces an image processing algorithm combining color and texture information, and the use of a support vector machine, to accelerate fruit detection by isolating useful features in images. Experiments carried out on two varieties of red grapes (Shiraz and Cabernet Sauvignon) demonstrate an accuracy of 87% and recall of 90%. This method is also shown to remove the restriction on the field of view and background, which limited existing methods and is a first step towards precise and reliable yield estimation on a large scale.

The objective for the present trial was to understand whether dairy heifers could be trained to respond to an audio cue paired with a feed reward. The use of acoustic conditioning to induce cattle movement has not previously been tested with animal-mounted devices to call cattle both individually and as a group. Five heifers underwent testing for 6 days as part of an 18-day field trial (12 days of conditioning). The 6-day testing and data-collection period involved the heifers being called via a smartphone device mounted on the cheek strap of a halter. Heifers were called either as individuals or as a group. When the audio cue was sent, heifers were expected to traffic from a group-holding area to a feeding area (~80-m distance) to receive an allocation of a grain-based concentrate. Heifers were significantly (P = 0.001) more likely to approach the feeding area when called as a group (91% response rate) than when they were called as individuals (67% response rate). When heifers did respond to being called, their time to traffic to the feed area was quicker (P < 0.001) when they were called as a group (77.9 ± 55.4 s) than when they were called as individuals (139.3 ± 89.2 s). The present trial has shown that animals can be trained to respond to an audio cue paired to a feed reward, highlighting the potential for acoustic conditioning to improve voluntary cow movement with an animal-mounted device. It also highlights the limitations of cattle responding to being called individually compared with being called as a group.

Kidnapping occurs when a robot is unaware that it has not correctly ascertained its position. As a result, the global map may be severely deformed and the robot may be unable to perform its function. This paper presents a metric-based technique for real-time kidnap detection that utilises a set of binary classifiers to identify all kidnapping events during the autonomous operation of a mobile robot. In contrast, existing techniques either solve specific cases of kidnapping, such as elevator motion, without addressing the general case or remove dependence on local pose estimation entirely, an inefficient and computationally expensive approach. Four metrics were evaluated and the optimal thresholds for the most suitable metrics were determined, resulting in a combined detector that has a negligible probability of failing to identify kidnapping events and a low false positive rate for both indoor and outdoor environments. While this paper uses metrics specific to 3D point clouds, the approach can be generalised to other forms of data, including visual, providing that two independent ways of estimating pose are available.

This paper presents an efficient approach to global path planning for multiple agents during large-scale map deformation. The problem of planning using dense data during large-scale map deformation is addressed by using a hybrid metric-topological planner that maintains locally consistent policies. These policies are cached, providing efficiency gains relative to alternate planning approaches that are characterized using complexity analysis. Simulation results show the effectiveness of this approach in handling notable map deformation while achieving good efficiency.

Ground plane detection plays an important role in stereo vision based obstacle detection methods. Recently, V-Disparity image has been widely used for ground plane detection. The existing approach based on V-Disparity image can detect flat ground successfully but have difficulty in detecting non-flat ground. In this paper, we discuss the representation of non-flat ground in V-Disparity image, based on which we propose a method to detect non-flat ground using V-Disparity image.

This paper presents a framework for efficient path planning in a deformable map. A roadmap and local cost maps are combined and integrated into a generic SLAM process to provide fast path querying for multiple sources and multiple destinations. Analysis of a simple deformation metric shows the ability of the framework to efficiently maintain a consistent plan during major map adjustment by updating the roadmap and selected local cost maps. Results from simulation verify the effectiveness of the framework in handling deformable maps in an efficient manner.

Komatsu Ltd. is reinforcing the R&D of automation technology for our earth-movers. Based on this trend, a path planning methodology for autonomous bulldozers is proposed and developed. This methodology autonomously plans an efficient path depending on a given material profile. Conventionally, existing path planning algorithms are versatile so they can be applied to any application, typically by using a grid-based map. However, in reality, a substantial effort is still required when applying to specific industry products. In contrast to this trend, the aim of this work is to develop a path planning algorithm specifically suitable for bulldozers, from the theory development phase. The novel planning methodology was developed by incorporating industry feedback, and has successfully resolved the issues which occurred when attempting to apply the existing planning methodologies. As a result of this work, the developed methodology provides an efficient path that lets bulldozers complete given tasks with minimal operation time, without a human operator on-board, and can be applied to commercial machines immediately.