A hidden Markov model and relative entropy rate approach to vision-based dim target detection for UAV sense-and-avoid

Uninhabited aerial vehicles (UAVs) are a cutting-edge technology that is at the forefront of aviation/aerospace research and development worldwide. Many consider their current military and defence applications as just a token of their enormous potential. Unlocking and fully exploiting this potential will see UAVs in a multitude of civilian applications and routinely operating alongside piloted aircraft. The key to realising the full potential of UAVs lies in addressing a host of regulatory, public relation, and technological challenges never encountered be- fore. Aircraft collision avoidance is considered to be one of the most important issues to be addressed, given its safety critical nature. The collision avoidance problem can be roughly organised into three areas: 1) Sense; 2) Detect; and 3) Avoid. Sensing is concerned with obtaining accurate and reliable information about other aircraft in the air; detection involves identifying potential collision threats based on available information; avoidance deals with the formulation and execution of appropriate manoeuvres to maintain safe separation. This thesis tackles the detection aspect of collision avoidance, via the development of a target detection algorithm that is capable of real-time operation onboard a UAV platform. One of the key challenges of the detection problem is the need to provide early warning. This translates to detecting potential threats whilst they are still far away, when their presence is likely to be obscured and hidden by noise. Another important consideration is the choice of sensors to capture target information, which has implications for the design and practical implementation of the detection algorithm. The main contributions of the thesis are: 1) the proposal of a dim target detection algorithm combining image morphology and hidden Markov model (HMM) filtering approaches; 2) the novel use of relative entropy rate (RER) concepts for HMM filter design; 3) the characterisation of algorithm detection performance based on simulated data as well as real in-flight target image data; and 4) the demonstration of the proposed algorithm's capacity for real-time target detection. We also consider the extension of HMM filtering techniques and the application of RER concepts for target heading angle estimation. In this thesis we propose a computer-vision based detection solution, due to the commercial-off-the-shelf (COTS) availability of camera hardware and the hardware's relatively low cost, power, and size requirements. The proposed target detection algorithm adopts a two-stage processing paradigm that begins with an image enhancement pre-processing stage followed by a track-before-detect (TBD) temporal processing stage that has been shown to be effective in dim target detection. We compare the performance of two candidate morphological filters for the image pre-processing stage, and propose a multiple hidden Markov model (MHMM) filter for the TBD temporal processing stage. The role of the morphological pre-processing stage is to exploit the spatial features of potential collision threats, while the MHMM filter serves to exploit the temporal characteristics or dynamics. The problem of optimising our proposed MHMM filter has been examined in detail. Our investigation has produced a novel design process for the MHMM filter that exploits information theory and entropy related concepts. The filter design process is posed as a mini-max optimisation problem based on a joint RER cost criterion. We provide proof that this joint RER cost criterion provides a bound on the conditional mean estimate (CME) performance of our MHMM filter, and this in turn establishes a strong theoretical basis connecting our filter design process to filter performance. Through this connection we can intelligently compare and optimise candidate filter models at the design stage, rather than having to resort to time consuming Monte Carlo simulations to gauge the relative performance of candidate designs. Moreover, the underlying entropy concepts are not constrained to any particular model type. This suggests that the RER concepts established here may be generalised to provide a useful design criterion for multiple model filtering approaches outside the class of HMM filters. In this thesis we also evaluate the performance of our proposed target detection algorithm under realistic operation conditions, and give consideration to the practical deployment of the detection algorithm onboard a UAV platform. Two fixed-wing UAVs were engaged to recreate various collision-course scenarios to capture highly realistic vision (from an onboard camera perspective) of the moments leading up to a collision. Based on this collected data, our proposed detection approach was able to detect targets out to distances ranging from about 400m to 900m. These distances, (with some assumptions about closing speeds and aircraft trajectories) translate to an advanced warning ahead of impact that approaches the 12.5 second response time recommended for human pilots. Furthermore, readily available graphic processing unit (GPU) based hardware is exploited for its parallel computing capabilities to demonstrate the practical feasibility of the proposed target detection algorithm. A prototype hardware-in- the-loop system has been found to be capable of achieving data processing rates sufficient for real-time operation. There is also scope for further improvement in performance through code optimisations. Overall, our proposed image-based target detection algorithm offers UAVs a cost-effective real-time target detection capability that is a step forward in ad- dressing the collision avoidance issue that is currently one of the most significant obstacles preventing widespread civilian applications of uninhabited aircraft. We also highlight that the algorithm development process has led to the discovery of a powerful multiple HMM filtering approach and a novel RER-based multiple filter design process. The utility of our multiple HMM filtering approach and RER concepts, however, extend beyond the target detection problem. This is demonstrated by our application of HMM filters and RER concepts to a heading angle estimation problem.

Image acquisition and manipulation protocols for CT-PET fusion to improve the accuracy of gross tumour volume localisation for 3D conformal radiotherapy for lung cancer

Aims: To develop clinical protocols for acquiring PET images, performing CT-PET registration and tumour volume definition based on the PET image data, for radiotherapy for lung cancer patients and then to test these protocols with respect to levels of accuracy and reproducibility. Method: A phantom-based quality assurance study of the processes associated with using registered CT and PET scans for tumour volume definition was conducted to: (1) investigate image acquisition and manipulation techniques for registering and contouring CT and PET images in a radiotherapy treatment planning system, and (2) determine technology-based errors in the registration and contouring processes. The outcomes of the phantom image based quality assurance study were used to determine clinical protocols. Protocols were developed for (1) acquiring patient PET image data for incorporation into the 3DCRT process, particularly for ensuring that the patient is positioned in their treatment position; (2) CT-PET image registration techniques and (3) GTV definition using the PET image data. The developed clinical protocols were tested using retrospective clinical trials to assess levels of inter-user variability which may be attributed to the use of these protocols. A Siemens Somatom Open Sensation 20 slice CT scanner and a Philips Allegro stand-alone PET scanner were used to acquire the images for this research. The Philips Pinnacle3 treatment planning system was used to perform the image registration and contouring of the CT and PET images. Results: Both the attenuation-corrected and transmission images obtained from standard whole-body PET staging clinical scanning protocols were acquired and imported into the treatment planning system for the phantom-based quality assurance study. Protocols for manipulating the PET images in the treatment planning system, particularly for quantifying uptake in volumes of interest and window levels for accurate geometric visualisation were determined. The automatic registration algorithms were found to have sub-voxel levels of accuracy, with transmission scan-based CT-PET registration more accurate than emission scan-based registration of the phantom images. Respiration induced image artifacts were not found to influence registration accuracy while inadequate pre-registration over-lap of the CT and PET images was found to result in large registration errors. A threshold value based on a percentage of the maximum uptake within a volume of interest was found to accurately contour the different features of the phantom despite the lower spatial resolution of the PET images. Appropriate selection of the threshold value is dependant on target-to-background ratios and the presence of respiratory motion. The results from the phantom-based study were used to design, implement and test clinical CT-PET fusion protocols. The patient PET image acquisition protocols enabled patients to be successfully identified and positioned in their radiotherapy treatment position during the acquisition of their whole-body PET staging scan. While automatic registration techniques were found to reduce inter-user variation compared to manual techniques, there was no significant difference in the registration outcomes for transmission or emission scan-based registration of the patient images, using the protocol. Tumour volumes contoured on registered patient CT-PET images using the tested threshold values and viewing windows determined from the phantom study, demonstrated less inter-user variation for the primary tumour volume contours than those contoured using only the patient’s planning CT scans. Conclusions: The developed clinical protocols allow a patient’s whole-body PET staging scan to be incorporated, manipulated and quantified in the treatment planning process to improve the accuracy of gross tumour volume localisation in 3D conformal radiotherapy for lung cancer. Image registration protocols which factor in potential software-based errors combined with adequate user training are recommended to increase the accuracy and reproducibility of registration outcomes. A semi-automated adaptive threshold contouring technique incorporating a PET windowing protocol, accurately defines the geometric edge of a tumour volume using PET image data from a stand alone PET scanner, including 4D target volumes.

Visual field size criteria for mobility rehabilitation referral

Purpose. To investigate evidence-based visual field size criteria for referral of low-vision (LV) patients for mobility rehabilitation. Methods. One hundred and nine participants with LV and 41 age-matched participants with normal sight (NS) were recruited. The LV group was heterogeneous with diverse causes of visual impairment. We measured binocular kinetic visual fields with the Humphrey Field Analyzer and mobility performance on an obstacle-rich, indoor course. Mobility was assessed as percent preferred walking speed (PPWS) and number of obstacle-contact errors. The weighted kappa coefficient of association (κr) was used to discriminate LV participants with both unsafe and inefficient mobility from those with adequate mobility on the basis of their visual field size for the full sample and for subgroups according to type of visual field loss and whether or not the participants had previously received orientation and mobility training. Results. LV participants with both PPWS <38% and errors >6 on our course were classified as having inadequate (inefficient and unsafe) mobility compared with NS participants. Mobility appeared to be first compromised when the visual field was less than about 1.2 steradians (sr; solid angle of a circular visual field of about 70° diameter). Visual fields <0.23 and 0.63 sr (31 to 52° diameter) discriminated patients with at-risk mobility for the full sample and across the two subgroups. A visual field of 0.05 sr (15° diameter) discriminated those with critical mobility. Conclusions. Our study suggests that: practitioners should be alert to potential mobility difficulties when the visual field is less than about 1.2 sr (70° diameter); assessment for mobility rehabilitation may be warranted when the visual field is constricted to about 0.23 to 0.63 sr (31 to 52° diameter) depending on the nature of their visual field loss and previous history (at risk); and mobility rehabilitation should be conducted before the visual field is constricted to 0.05 sr (15° diameter; critical).

Unaided stereo vision based pose estimation

This paper presents the development of a low-cost sensor platform for use in ground-based visual pose estimation and scene mapping tasks. We seek to develop a technical solution using low-cost vision hardware that allows us to accurately estimate robot position for SLAM tasks. We present results from the application of a vision based pose estimation technique to simultaneously determine camera poses and scene structure. The results are generated from a dataset gathered traversing a local road at the St Lucia Campus of the University of Queensland. We show the accuracy of the pose estimation over a 1.6km trajectory in relation to GPS ground truth.

Developments in texture-based vector field visualisation techniques

Vector field visualisation is one of the classic sub-fields of scientific data visualisation. The need for effective visualisation of flow data arises in many scientific domains ranging from medical sciences to aerodynamics. Though there has been much research on the topic, the question of how to communicate flow information effectively in real, practical situations is still largely an unsolved problem. This is particularly true for complex 3D flows. In this presentation we give a brief introduction and background to vector field visualisation and comment on the effectiveness of the most common solutions. We will then give some examples of current development on texture-based techniques, and given practical examples of their use in CFD research and hydrodynamic applications.

Omnidirectional bearing-only see-and-avoid for small aerial robots

Motivated by the growing interest in unmanned aerial system’s applications in indoor and outdoor settings and the standardisation of visual sensors as vehicle payload. This work presents a collision avoidance approach based on omnidirectional cameras that does not require the estimation of range between two platforms to resolve a collision encounter. It will achieve a minimum separation between the two vehicles involved by maximising the view-angle given by the omnidirectional sensor. Only visual information is used to achieve avoidance under a bearing-only visual servoing approach. We provide theoretical problem formulation, as well as results from real flight using small quadrotors.

Autonomous aerial navigation and tracking of marine animals

In this paper, we describe the development of an independent and on-board visual servoing system which allows a computationally impoverished aerial vehicle to autonomously identify and track a moving surface target. Our image segmentation and target identification algorithms were developed with the specific task of monitoring whales at sea but could be adapted for other targets. Observing whales is important for many marine biology tasks and is currently performed manually from the shore or from boats. We also present hardware experiments which demonstrate the capabilities of our algorithms for object identification and tracking that enable a flying vehicle to track a moving target.

See-and-avoid quadcopter using fuzzy control optimized by cross-entropy

In this work we present an optimized fuzzy visual servoing system for obstacle avoidance using an unmanned aerial vehicle. The cross-entropy theory is used to optimise the gains of our controllers. The optimization process was made using the ROS-Gazebo 3D simulation with purposeful extensions developed for our experiments. Visual servoing is achieved through an image processing front-end that uses the Camshift algorithm to detect and track objects in the scene. Experimental flight trials using a small quadrotor were performed to validate the parameters estimated from simulation. The integration of cross- entropy methods is a straightforward way to estimate optimal gains achieving excellent results when tested in real flights.

UAS See-and-Avoid using two different approaches of Fuzzy Control

This work presents two UAS See and Avoid approaches using Fuzzy Control. We compare the performance of each controller when a Cross-Entropy method is applied to optimase the parameters for one of the controllers. Each controller receive information from an image processing front-end that detect and track targets in the environment. Visual information is then used under a visual servoing approach to perform autonomous avoidance. Experimental flight trials using a small quadrotor were performed to validate and compare the behaviour of both controllers

Fusion of morphological images for airborne target detection

Several track-before-detection approaches for image based aircraft detection have recently been examined in an important automated aircraft collision detection application. A particularly popular approach is a two stage processing paradigm which involves: a morphological spatial filter stage (which aims to emphasize the visual characteristics of targets) followed by a temporal or track filter stage (which aims to emphasize the temporal characteristics of targets). In this paper, we proposed new spot detection techniques for this two stage processing paradigm that fuse together raw and morphological images or fuse together various different morphological images (we call these approaches morphological reinforcement). On the basis of flight test data, the proposed morphological reinforcement operations are shown to offer superior signal to-noise characteristics when compared to standard spatial filter options (such as the close-minus-open and adaptive contour morphological operations). However, system operation characterised curves, which examine detection verses false alarm characteristics after both processing stages, illustrate that system performance is very data dependent.

Visualisation of complex flows using texture-based techniques

Complex flow datasets are often difficult to represent in detail using traditional vector visualisation techniques such as arrow plots and streamlines. This is particularly true when the flow regime changes in time. Texture-based techniques, which are based on the advection of dense textures, are novel techniques for visualising such flows (i.e., complex dynamics and time-dependent). In this paper, we review two popular texture-based techniques and their application to flow datasets sourced from real research projects. The texture-based techniques investigated were Line Integral Convolution (LIC), and Image-Based Flow Visualisation (IBFV). We evaluated these techniques and in this paper report on their visualisation effectiveness (when compared with traditional techniques), their ease of implementation, and their computational overhead.

Visualisation of complex flows using texture-based techniques

Detailed representations of complex flow datasets are often difficult to generate using traditional vector visualisation techniques such as arrow plots and streamlines. This is particularly true when the flow regime changes in time. Texture-based techniques, which are based on the advection of dense textures, are novel techniques for visualising such flows. We review two popular texture based techniques and their application to flow datasets sourced from active research projects. The techniques investigated were Line integral convolution (LIC) [1], and Image based flow visualisation (IBFV) [18]. We evaluated these and report on their effectiveness from a visualisation perspective. We also report on their ease of implementation and computational overheads.

High altitude stereo visual odometry

Stereo visual odometry has received little investigation in high altitude applications due to the generally poor performance of rigid stereo rigs at extremely small baseline-to-depth ratios. Without additional sensing, metric scale is considered lost and odometry is seen as effective only for monocular perspectives. This paper presents a novel modification to stereo based visual odometry that allows accurate, metric pose estimation from high altitudes, even in the presence of poor calibration and without additional sensor inputs. By relaxing the (typically fixed) stereo transform during bundle adjustment and reducing the dependence on the fixed geometry for triangulation, metrically scaled visual odometry can be obtained in situations where high altitude and structural deformation from vibration would cause traditional algorithms to fail. This is achieved through the use of a novel constrained bundle adjustment routine and accurately scaled pose initializer. We present visual odometry results demonstrating the technique on a short-baseline stereo pair inside a fixed-wing UAV flying at significant height (~30-100m).

Visual metrics for the evaluation of sensor data quality in outdoor perception

This paper proposes an experimental study of quality metrics that can be applied to visual and infrared images acquired from cameras onboard an unmanned ground vehicle (UGV). The relevance of existing metrics in this context is discussed and a novel metric is introduced. Selected metrics are evaluated on data collected by a UGV in clear and challenging environmental conditions, represented in this paper by the presence of airborne dust or smoke. An example of application is given with monocular SLAM estimating the pose of the UGV while smoke is present in the environment. It is shown that the proposed novel quality metric can be used to anticipate situations where the quality of the pose estimate will be significantly degraded due to the input image data. This leads to decisions of advantageously switching between data sources (e.g. using infrared images instead of visual images).

Outdoor flight testing of a pole inspection UAV incorporating high-speed vision

We present a pole inspection system for outdoor environments comprising a high-speed camera on a vertical take-off and landing (VTOL) aerial platform. The pole inspection task requires a vehicle to fly close to a structure while maintaining a fixed stand-off distance from it. Typical GPS errors make GPS-based navigation unsuitable for this task however. When flying outdoors a vehicle is also affected by aerodynamics disturbances such as wind gusts, so the onboard controller must be robust to these disturbances in order to maintain the stand-off distance. Two problems must therefor be addressed: fast and accurate state estimation without GPS, and the design of a robust controller. We resolve these problems by a) performing visual + inertial relative state estimation and b) using a robust line tracker and a nested controller design. Our state estimation exploits high-speed camera images (100Hz) and 70Hz IMU data fused in an Extended Kalman Filter (EKF). We demonstrate results from outdoor experiments for pole-relative hovering, and pole circumnavigation where the operator provides only yaw commands. Lastly, we show results for image-based 3D reconstruction and texture mapping of a pole to demonstrate the usefulness for inspection tasks.