Robust detection, localization, and tracking of objects are essential for autonomous driving. Computer vision has largely driven development based on camera sensors in recent years, but 3D localization from images is still challenging. Sensors such as LiDAR or RADAR are used to compute depth; each having its own advantages and drawbacks. The main idea of the project is to be able to mix images from the camera and RADAR detections in order to estimate depths for the objects appearing in the images. Fusion strategies can be considered the solution to give a more detailed description of the environment by utilizing both the 3D localization capabilities of range sensors and the higher spatial resolution of image data. The idea is to fuse 3D detections from the RADAR onto the image plane, this requires a high level of synchronization of the sensors and projections of the RADAR data on the required image. / Robust detektering, lokalisering och spårning av objekt är avgörande för autonom körning. Datorseende har till stor del drivit utvecklingen baserad på kamerasensorer de senaste åren, men 3D-lokalisering från bilder är fortfarande utmanande. Sensorer som LiDAR eller RADAR används för att beräkna djup; var och en har sina egna fördelar och nackdelar. Huvudtanken med projektet är att kunna blanda bilder från kameran och RADAR-detektioner för att uppskatta djup för de objekt som förekommer i bilderna. Fusionsstrategier kan anses vara lösningen för att ge en mer detaljerad beskrivning av miljön med både 3D-lokaliseringsförmågan hos avståndssensorer och den högre rumsliga upplösningen av bilddata. Tanken är att smälta samman 3D-detektioner från RADAR till bildplanet, detta kräver en hög nivå av synkronisering av sensorerna och projektioner av RADAR-data på den önskade bilden.

When a Time-of-Flight (ToF) depth camera is used to monitor a region of interest, it has to be mounted correctly and have information regarding its position. Manual configuration currently require managing captured 3D ToF data in a 2D environment, which limits the user and might give rise to errors due to misinterpretation of the data. This thesis investigates if a real time 3D reconstruction mesh from a Microsoft HoloLens can be used as a target for point cloud registration using the ToF data, thus configuring the camera autonomously. Three registration algorithms, Fast Global Registration (FGR), Joint Registration Multiple Point Clouds (JR-MPC) and Prerejective RANSAC, were evaluated for this purpose. It was concluded that despite using different sensors it is possible to perform accurate registration. Also, it was shown that the registration can be done accurately within a reasonable time, compared with the inherent time to perform 3D reconstruction on the Hololens. All algorithms could solve the problem, but it was concluded that FGR provided the most satisfying results, though requiring several constraints on the data.

空載光達系統於建置國土測繪基本資料扮演關鍵角色，依國土測繪法，為確保測繪成果品質，應依測量計畫目的及作業精度需求辦理儀器校正。國土測繪中心已於102年度建置航遙測感應器系統校正作業中，提出矩形建物之平屋頂面做為空載光達系統校正之可行性，而其所稱之校正，是以點雲精度評估待校件空載光達系統所得最終成果品質，並不對儀器做任何參數改正，但其校正成果可能因不同人員操作而有差異，因此本研究嘗試建立一套空載光達點雲半自動化精度評估程序，此外探討以山形屋脊線執行點雲精度評估之可行性。 由於光達點雲為離散的三維資訊，不論是以山形屋脊線或矩形建物之平屋頂面作為標物執行點雲精度評估，均須先萃取屋頂面上之點，為避免萃取成果受雜訊影響，本研究引入粗差偵測理論，發展最小一乘法結合李德仁以後驗變方估計原理導出的選擇權迭代法(李德仁法)將非屋頂點視為粗差排除。研究中分別對矩形建物之平屋頂面及山形屋脊線進行模擬及真實資料實驗，其中山形屋脊線作為點雲精度評估之可行性實驗中發現不適合用於評估點雲精度，因此後續實驗僅以萃取矩形建物之平屋頂面點雲過程探討粗差比率對半自動化點雲精度評估程序之影響。模擬實驗成果顯示最小一乘法有助於提升李德仁法偵測粗差數量5%至10%；真實資料實驗，以含有牆面點雲的狀況為例，則有助提升5%的偵測粗差數量。本研究由逐步測試結果提出能夠適用於真實狀況的半自動化之點雲精度評估程序，即使由不同人員操作，仍能獲得一致的成果，顯示本研究半自動化精度評估程序之可信度。 / The airborne LiDAR system plays a crucial role in building land surveying data. Based on the Land Surveying and Mapping Act, to ensure the quality of surveying, instrument calibration is required. The approach proposed by National Land Surveying and Mapping Center (NLSC) in 2013 was confirmed the feasibility for airborne LiDAR system calibration using rectangular horizontal roof plane. The calibration mean to assess the final quality of airborne LiDAR system based on the assessment of the accuracy of the point cloud, and do not adjust the instrument. But the results may vary according to different operators. This study attempts to establish a semi-automatic procedure for the accuracy assessment of point clouds from airborne LiDAR system. In addition, the gable roof ridge lines is discussed for its feasibility for the accuracy assessment of point cloud. No matter that calibration is performed using rectangular horizontal roof plane or gable roof ridge line, point clouds located on roof planes need to be extracted at first. Therefore, Least Absolute Deviation (LAD) combined with the Iteration using Selected Weights (Deren Li method) is developed to exclude the non-roof points which regarded as gross errors and eliminate their influences. The simulated test and actual data test found that gable roof ridge lines are not suitable for accuracy assessment. As for the simulated test using horizontal roof planes, LAD combined with Deren Li method prompts the rate of gross error detection about 5% to 10% than that only by Deren Li method. In actual test, data contains wall points, LAD combined with Deren Li method can prompt about 5%. Meanwhile, a semi-automatic procedure for real operations is proposed by the step-by-step test. Even different operators employ this semi-automatic procedure, consistent results will be obtained and the reliability can achieve.

Cette thèse présente trois contributions en lien avec l'apprentissage et la détection de dissimilarité statistique multivariée, problématique d'importance primordiale pour de nombreuses méthodes d'apprentissage utilisées dans un nombre croissant de domaines. La première contribution introduit la notion de taille d'effet multivariée non-paramétrique, éclairant la nature de la dissimilarité détectée entre deux jeux de données, en deux étapes. La première consiste en une décomposition d'une mesure de dissimilarité (divergence de Jensen-Shannon) visant à la localiser dans l'espace ambiant, tandis que la seconde génère un résultat facilement interprétable en termes de grappes de points de forte discrépance et en proximité spatiale. La seconde contribution présente le premier test non-paramétrique d'homogénéité séquentiel, traitant les données issues de deux jeux une à une--au lieu de considérer ceux-ci- in extenso. Le test peut ainsi être arrêté dès qu'une évidence suffisamment forte est observée, offrant une flexibilité accrue tout en garantissant un contrôle del'erreur de type I. Sous certaines conditions, nous établissons aussi que le test a asymptotiquement une probabilité d'erreur de type II tendant vers zéro. La troisième contribution consiste en un test de détection de changement séquentiel basé sur deux fenêtres glissantes sur lesquelles un test d'homogénéité est effectué, avec des garanties sur l'erreur de type I. Notre test a une empreinte mémoire contrôlée et, contrairement à des méthodes de l'état de l'art qui ont aussi un contrôle sur l'erreur de type I, a une complexité en temps constante par observation, le rendant adapté aux flux de données. / In this thesis, we study problems related to learning and detecting multivariate statistical dissimilarity, which are of paramount importance for many statistical learning methods nowadays used in an increasingly number of fields. This thesis makes three contributions related to these problems. The first contribution introduces a notion of multivariate nonparametric effect size shedding light on the nature of the dissimilarity detected between two datasets. Our two step method first decomposes a dissimilarity measure (Jensen-Shannon divergence) aiming at localizing the dissimilarity in the data embedding space, and then proceeds by aggregating points of high discrepancy and in spatial proximity into clusters. The second contribution presents the first sequential nonparametric two-sample test. That is, instead of being given two sets of observations of fixed size, observations can be treated one at a time and, when strongly enough evidence has been found, the test can be stopped, yielding a more flexible procedure while keeping guaranteed type I error control. Additionally, under certain conditions, when the number of observations tends to infinity, the test has a vanishing probability of type II error. The third contribution consists in a sequential change detection test based on two sliding windows on which a two-sample test is performed, with type I error guarantees. Our test has controlled memory footprint and, as opposed to state-of-the-art methods that also provide type I error control, has constant time complexity per observation, which makes our test suitable for streaming data.