Automatisk travmätning är ett mätsystem som mäter vedvolymen på virkeslastbilar. Systemet består av sex stycken sensor-system. Varje sensor kalibreras först individuellt och sedan ihop för att ge ett sammanfogat världskoordinat system. Varje sensor genererar en djupbild och en reflektansbild, där värdena i djupbilden representerar avståndet från kameran. Uppdragsgivaren har utvecklat en algoritm som utifrån mätdatat(bilderna)  uppskattar vedvolymen till en viss noggrannhet som uppfyller kraven ställda av skogsindustrin för automatisk mätning av travar på virkeslastbil. I den här rapporten undersöks om bättre mätresultat kan uppnås exempelvis med andra metoder eller kombinationer av dem.Till förfogande finns ca 125 dataset av travar där facit finns. Facit består av manuella stickprovsmätningar där varje enskild stock mätts för sig. Initialt valdes aktivt att inte sätta sig in i uppdragsgivarens algoritm för att inte bli färgad av hur de kommit fram till sina resultat. Främst används fram- och baksidebilderna av entrave för att hitta stockarna. Därefter interpoleras de funna stockarna in till mitten av traven eller så paras stockarna ihop från de båda sidorna. Ibland finns vissa problem med bilderna. Oftast är minst en av sidorna ockluderade av lastbilshytten, kranen eller en annan trave. Då gäller det att hitta uppskattning utifrån det data man ser för fylla upp de skymda områdena.I början av examensarbetet användes två metoder(MSER och Punktplanmetoden) för undersöka om man kunde uppnå bra resultat utifrån att enbart mäta datat och användadet som initial gissning till volymen. Dock upptäcktes det att värdefulla detaljer   i dataseten missades för att mer noggrant bestämma vedvolymen. Exempel på sådan data är fördelningen av diametern på de funna stockändarna. Tillika tenderades kraftig överestimering när travarna innehöll en viss mängd ris och eller dåligt kvistade stockar. Därefter konstruerades en geometrisk metod, och det var den här metoden som det lades mest tid på.I figurerna nedan visas en tabell och en graf där alla tre metoders resultat under bark(UB) visas och intervall gränserna för att uppfylla kraven ställda av skogsindustrin. / Automatic measurment of volume for on-truck timber stacks is a measurment system that measures the woodvolume. The whole system consist of six sensorsystems. Each sensor is at first individually calibrated and then calibrated together to give a merged worldcoordinate system. Each sensor produces a depth image and are flectance image where the values in the depth image represents the distance from the sensor. The consistuent has developed a algoritm which from the measured data estimates the woodvolume to a certain accuracy that fullfills the requirements stated from the forest industry for automatic measurent of timber stacks on trucks. I this thesis we will analyze if better measurement results can be achieved with other methods or combinations of them. At disposal we have about 125 datasets of timber stacks where the key exists. The key consists of random samples that have been manually measured log by log. A active choice to not take part of the consistuent´s algoritm was made to not be coloured of how they acheived their results. Mainly the front- and backside images of at imberstack is used to find the logs. Subsequentlythefound logs are interpolated into the middle of the timber stack or are paired together. Sometimes there exists some problems with the images. Frequently at least on the sides are occluded by the cabin, crane or another timber stack. Then estimation is required from the visible data to make a qualified guess of the occluded areas. In the beginning of this thesis two methods where analysed, the MSER features method and the pointplane method. The main purpose of those two methods was to analyse if a good results could be acheived by just measuring the data and using it as a initial volume estimate. Though it was discovered that valuable details was missed to be able to more accuratly determine the woodvolume. Examples of such details is the distribution of diameter on the found logs. Also these methods tended to heavily over- or underestimate when the timber stack had a certain amount of brushwood and or badly limbed logs. After that a geometric method was constructed, and it was this method that most time and analysis was spent on. In the figures below a table and a graph where the under bark results of all three methods and the interval limits of forest industry requirements.

This thesis presents an innovative decentralised sensor fusion framework with significant potential to improve navigation accuracy in autonomous vehicles. Its applicability is especially noteworthy in demanding scenarios, such as adverse weather conditions and intricate urban environments. In general, sensor fusion is a crucial method for integrating signals from various sources, extracting and integrating information from multiple inputs into a unified signal or data set. Frequently, sources of information are from sensors or devices designed for the perception and measurement of dynamic environmental changes. The collected data from diverse sensors undergoes processing through specialised algorithms, commonly referred to as "sensor fusion" or "data fusion" algorithms. This thesis describes sensor fusion's significance in processing data from multiple sources. It highlights the classification of fusion algorithms, demonstrating the versatility and applicability of sensor fusion across a range of redundant sensors. Moreover, various creative strategies for sensor fusion, including fault detection and isolation and methods for addressing non-Gaussian noise through smoothing filter techniques, are collectively introduced as part of a comprehensive navigation framework. The contributions of this thesis are summarized in the following. First, it introduces a decentralised two-layered fusion architecture for pose estimation, emphasising fault resilience. In a decentralised fashion, it utilises distributed nodes equipped with extended Kalman filters in the initial tier and optimal information filters in the subsequent tier to amalgamate pose data from multiple sensors. The design is named the Fault-Resilient Optimal Information Fusion (FR-OIF) architecture in this thesis, which guarantees reliable pose estimation, even in cases of sensor malfunctions. Secondly, this work proposes an Auto-encoder-based fault detection framework for a multi-sensorial distributed pose estimation. In this framework, auto-encoders are applied to detect anomalies in the raw signal measurements. At the same time, a fault-resilient optimal information filter (FROIF) approach is incorporated with the auto-encoder-based detection to improve estimation accuracy. The effectiveness of these methods is demonstrated through experimental results involving a micro aerial vehicle and is compared to a novel classical detection approach based on the Extended Kalman filter. Furthermore, it introduces an integrated multi-sensor fusion architecture enhanced by centralised Auto-encoder technology and an EKF framework. This approach effectively removes sensor data noise and anomalies, ensuring reliable data reconstruction, even when faced with time-dependent anomalies. The assessment of the framework's performance using actual sensor data collected from the onboard sensors of a micro aerial vehicle demonstrates its superiority compared to a centralised Extended Kalman filter without Auto-encoders. The next part of the thesis discusses the increasing need for resilient autonomy in complex space missions. It emphasises the challenges posed by interactions with non-cooperative objects and extreme environments, calling for advanced autonomy solutions.  Furthermore, this work introduces a decentralised multi-sensor fusion architecture for resilient satellite navigation around asteroids. It addresses challenges such as dynamic illumination, sensor drift, and momentary sensor failure. The approach includes fault detection and isolation methods, ensuring autonomous operation in adverse conditions. Finally, the last part of the thesis focuses on accurate localisation and deviation identification in multi-sensor fusion with Millimeter-Wave Radars. It presents a flexible, decentralised smoothing filter framework that effectively handles unwanted measurements and enhances Ego velocity estimation accuracy.  Overall, this thesis plays a significant role in advancing the field of decentralised sensor fusion, encompassing anomaly avoidance mechanisms, fault detection and isolation frameworks, and robust navigation algorithms applicable across a range of domains, covering everything from robotics to space exploration. In the initial section of this thesis, we delve into the backdrop, reasons behind the research, existing challenges, and the contributions made. Conversely, the subsequent section comprises the complete articles linked to the outlined contributions and a bibliography.

This bachelor thesis investigates the possibility of implementing a system to detect defects with extractable data from ABB’s servo guns. By looking at the deflection data i.e., the electrode position at every stroke, multiple methods are applied to investigate the behavior of the data. The distribution of the data that was first investigated led the thesis straight to one of the traditional methods in Statistical Process Control (SPC) called Control Chart. After investigating the data’s frequency domain, pattern and applying unsupervised clustering, several experiments were performed. The first experiment was to see how the deflection changed due to a change in force applied between the electrode tips. Second, artificial defects was created by switching electrode tips of different lengths and change the set force between the electrode tips. The idea behind the experiments was to evaluate the performance of a control chart whereas the lower and upper control levels was based on a regression curve that was implemented by a set force for a specific servo gun. The method proved to work for the tests that were performed, but as the results suggest it needs more research and development to be suitable for implementation in ABB’s software. / Detta examensarbete undersöker möjligheten att implementera ett system som detekterar defekter m.h.a. böjningsdata från ABB’s svetstänger. Genom att titta på datan, dvs., elektrodernas position vid varje slag/svets, används flertal metoder för att undersöka processens beteende. Datans distribution visade en god passform till en normalfördelning, vilket ledde projektet till en traditionell metod inom Statistical Process Control (SPC) sk. Control Chart. Efter att man undersökt datans frekvensdomän, om det fanns ett mönster att avläsa och tillämpat Unsupervised clustering, genomfördes flera olika experiment. Det första experimentet undersökte hur böjningen i tångarmarna förändrades genom att applicera olika krafter mellan elektroderna. Dessutom genererades data för att försöka skapa konstgjorda sprickor genom att byta ut elektroder av olika längd. I tillägg skapades drivande data (förändringar i böjning med tiden) genom att applicera små förändringar i kraft mellan elektroderna i ett visst intervall. Idén bakom experimenten var att utvärdera prestationen av en Control Chart, där övre och undre kontrollparametrar bestämdes utifrån en regressionsfunktion som var implementerad m.h.a. parametrar från resultatet utav förhållandet mellan kraft och böjning. Experimenten utfördes på en utav ABB’s svettänger s.k. GWT X9. Metoden visade sig fungera för testen som blev utförda, men resultatet föreslår att arbetet kräver mer foskning och arbete för att bli implementerad i ABB’s mjukvara.

this thesis the problem of maximum likelihood (ML) detection for the linear multiple-input multiple-output (MIMO) channel is considered. The thesis investigates two algorithms previously proposed in the literature for implementing the ML detector, namely semide nite relaxation and sphere decoding. The first algorithm, semide nite relaxation, is a suboptimal implementation of the ML detector meaning that it is not guaranteed to solve the maximum likelihood detection problem. Still, numerical evidence suggests that the performance of the semide nite relaxation detector is close to that of the true ML detector. A contribution made in this thesis is to derive conditions under which the semide nite relaxation estimate can be guaranteed to coincide with the ML estimate. The second algorithm, the sphere decoder, can be used to solve the ML detection problem exactly. Numerical evidence has previously shown that the complexity of the sphere decoder is remarkably low for problems of moderate size. This has led to the widespread belief that the sphere decoder is of polynomial expected complexity. This is however unfortunately not true. Instead, in most scenarios encountered in digital communications, the expected complexity of the algorithm is exponential in the number of symbols jointly detected. However, for high signal to noise ratio the rate of exponential increase is small. In this thesis it is proved that for a large class of detection problems the expected complexity is lower bounded by an exponential function. Also, for the special case of an i.i.d. Rayleigh fading channel, an asymptotic analysis is presented which enables the computation of the expected complexity up to the linear term in the exponent.

Two estimation problems are treated in this thesis. Estimators are suggested and the asymptotical properties of the estimates are investigated analytically. Numerical simulations are used to assess small-sample performance. In addition, performance bounds are calculated. The first problem treated is parameter estimation for the reduced rank linear regression. A new method based on instrumental variable principles is proposed and its asymptotical performance analyzed. In addition, the Cram\'r-Rao bound for the problem is derived for a general Gaussian noise model. The new method is asymptotically efficient (it has the smallest possible covariance) if the noise is temporally white, and outperforms previously suggested algorithms when the noise is temporally correlated. The approximation of a matrix with one of lower rank under a weighted norm is needed as part of the estimation algorithm. Two new, computationally efficient, methods are suggested. While the general matrix approximation problem has no known closed form solution, the proposed methods are asymptotically optimal as part of the estimation procedure in question. A new algorithm is also suggested for the related rank detection problem. The second part of this thesis treats direction of arrival estimation for narrowband signals using an array of sensors. Most algorithms require the noise covariance matrix to be known (up to a scaling factor) or to possess a known structure. In many cases the noise covariance is in fact estimated from a separate batch of signal-free samples. This work addresses the combined effects of finite sample sizes both in the estimated noise covariance matrix and in the data with signals present. No assumption is made on the structure of the noise covariance. The asymptotical covariance of weighted subspace fitting (WSF) is derived for the case in which the data are whitened using the noise covariance estimate. The obtained expression suggests an optimal weighting that improves performance compared to the standard choice. In addition, a new method based on covariance matching is proposed. Both methods are asymptotically statistically efficient. The Cramér-Rao bound for the problem is derived, and the expression becomes surprisingly simple. / <p>QC 20110103</p>

Multi-robot systems have important applications, such as space explorations, underwater missions, and surveillance operations. In most of these cases robots need to exchange data through communication. Limitations in the communication system however impose constraints on the design of coordination strategies. In this thesis we present three papers on cooperative control problems in which different communication constraints are considered. The first paper describes a rendezvous problem for a team of robots that exchanges position information through communication. A local control law for each robot should steer the team to a common meeting point when communicated data are quantized. The robots are not equipped with any sensors so the positions of other teammates are not measured. Two different types of quantized communication are considered: uniform and logarithmic. Logarithmic quantization is often preferable since it requires that fewer bits are communicated compared to when uniform quantization is used. For a class of feasible communication topologies, control laws that solve the rendezvous problem are derived. A hierarchical control structure is proposed in the second paper, for modelling autonomous underwater vehicles employed in finding a minimum of a scalar field. The controller is composed of two layers. The upper layer is the team controller, which is modeled as discrete-event system. It generates waypoints based on the simplex search optimization algorithm. The waypoints are used as target points by the lower control layer, which continuously steers each vehicle from the current to the next waypoint. It is shown that the communication of measurements is needed at each step for the team controller to generate unique waypoints. A protocol is proposed to reduce the amount of data to be exchanged, motivated by that underwater communication is costly in terms of energy. In the third paper, a probabilistic pursuit{evasion game is considered as an example to study constrained communication in multi-robot systems. This system can be used to model search-and-rescue operations and multi-robot exploration. Communication protocols based on time-triggered and event-triggered synchronization schemes are considered. It is shown that by limiting the communication to events when the probabilistic map updated by the individual pursuer contains new information, as measured by a map entropy, the utilization of the communication link can be considerably improved compared to conventional time-triggered communication.

This thesis begins with a survey of current state-of-the art in-car navigation systems. The pros and cons of the four commonly used information sources — GNSS/RF-based positioning, vehicle motion sensors, vehicle models and map information — are described. Common filters to combine the information from the various sources are discussed. Next, a GNSS-aided inertial navigation platform is presented, into which further sensors such as a camera and wheel-speed encoder can be incorporated. The construction of the hardware platform, together with an extended Kalman filter for a closed-loop integration between the GNSS receiver and the inertial navigation system (INS), is described. Results from a field test are presented. Thereafter, an approach is studied for calibrating a low-cost inertial measurement unit (IMU), requiring no mechanical platform for the accelerometer calibration and only a simple rotating table for the gyro calibration. The performance of the calibration algorithm is compared with the Cramér-Rao bound for cases where a mechanical platform is used to rotate the IMU into different precisely controlled orientations. Finally, the effects of time synchronization errors in a GNSS-aided INS are studied in terms of the increased error covariance of the state vector. Expressions for evaluating the error covariance of the navigation state vector are derived. Two different cases are studied in some detail. The first considers a navigation system in which the timing error is not taken into account by the integration filter. This leads to a system with an increased error covariance and a bias in the estimated forward acceleration. In the second case, a parameterization of the timing error is included as part of the estimation problem in the data integration. The estimated timing error is fed back to control an adjustable fractional delay filter, synchronizing the IMU and GNSS-receiver data. / QC 20101117

Measuring construction tool activity has a potential to improve tool productivity, reduce down-time and give insights to various construction processes. Today, a lot of data is being recordedfrom a construction site. This research aims to explore the technical feasibility of handheld powertool activity recognition with real-time tri-axial accelerometer data. The present study has threefocus areas: 1) Data collection using real-time accelerometer data from two tools: a combihammerand a screwdriver. 2) Hand-craft time and frequency domain features from the collected data. 3)Develop two classification algorithms, namely decision trees and random forest, with hand craftedfeatures to detect tool usage activities. The hand-crafted features provide an understanding of themechanical properties of the tools. For the combihammer, the activities recognized were hammerdrilling, chiseling and motor stop. The activity recognition accuracy was 79% with a decision treeand 80.8% with a random forest. For the screwdriver, the activities recognized were screwing,unscrewing and motor stop. The activity recognition accuracy was 87.7% for a decision tree and94.5% for a random forest algorithm. Variance from time domain and energy from frequencydomain were detected as the high importance features by both the classification algorithms forboth the tools.

Real-time imaging can be used to monitor, analyze and control medical treatments. In this thesis, we want to explain the spatiotemporal motion and thus enable more advanced procedures, especially real-time adaptation in radiation therapy. The motion occurring between image acquisitions can be quantified by image registration, which generates a mapping between the images. The contribution of the thesis consists of three papers, where we have used different approaches to estimate the motion between images. In Paper I, we combine a state-of-the-art method in real-time tracking with a learned sparse-to-dense interpolation scheme. For this, we track an arbitrary number of regions in a sequence of medical images. We estimated a sparse displacement field, based on the tracking positions and used the interpolation network to achieve its dense representation. Paper II was a contribution to a challenge in learnable image registration where we finished at 2nd place. Here we train a deep learning method to estimate the dense displacement field between two images. For this, we used a network architecture inspired by both conventional medical image registration methods and optical flow in computer vision. For Paper III, we estimate the dynamics of spatiotemporal images by training a generative network. We use nonlinear dimensional reduction techniques and assume a linear dynamic in a low-dimensional latent space. In comparison with conventional image registration methods, we provide a method more suitable for real-world scenarios, with the possibility of imputation and extrapolation. Although the problem is challenging and several questions are left unanswered we believe a combination of conventional, learnable, and dynamic modeling of the motion is the way forward.

Listeners with normal-hearing often overlook their ability to comprehend speech in noisy environments effortlessly. Our brain’s adeptness at identifying and amplifying attended voices while suppressing unwanted background noise, known as the cocktail party problem, has been extensively researched for decades. Yet, many aspects of this complex puzzle remain unsolved and listeners with hearing-impairment still struggle to focus on a specific speaker in noisy environments. While recent intelligent hearing aids have improved noise suppression, the problem of deciding which speaker to enhance remains unsolved, leading to discomfort for many hearing aid users in noisy environments. In this thesis, we explore the complexities of the human brain in challenging auditory environments. Two datasets are investigated where participants were tasked to selectively attend to one of two competing voices, replicating a cocktail-party scenario. The auditory stimuli trigger neurons to generate electrical signals that propagate in all directions. When a substantial number of neurons fire simultaneously, their collective electrical signal becomes detectable by small electrodes placed on the head. This method of measuring brain activity, known as electroencephalography (EEG), holds potential to provide feedback to the hearing aids, enabling adjustments to enhance attended voice(s). EEG data is often noisy, incorporating neural responses with artifacts such as muscle movements, eye blinks and heartbeats. In the first contribution of this thesis, we focus on comparing different manual and automatic artifact-rejection techniques and assessing their impact on auditory attention decoding (AAD). While EEG measurements offer high temporal accuracy, spatial resolution is inferior compared to alternative tools like magnetoencephalography (MEG). This difference poses a considerable challenge for source localization with EEG data. In the second contribution of this thesis, we demonstrate anticipated activity in the auditory cortex using EEG data from a single listener, employing Neuro-Current Response Functions (NCRFs). This method, previously evaluated only with MEG data, holds significant promise in hearing aid development. EEG data may involve both linear and nonlinear components due to the propagation of the electrical signals through brain tissue, skull, and scalp with varying conductivities. In the third contribution, we aim to enhance source localization by introducing a binning-based nonlinear detection and compensation method. The results suggest that compensating for some nonlinear components produces more precise and synchronized source localization compared to original EEG data. In the fourth contribution, we present a novel domain adaptation framework that improves AAD performances for listeners with initially low classification accuracy. This framework focuses on classifying the direction (left or right) of attended speech and shows a significant accuracy improvement when transporting poor data from one listener to the domain of good data from different listeners. Taken together, the contributions of this thesis hold promise for improving the lives of hearing-impaired individuals by closing the loop between the brain and hearing aids.