Robust UAV Localization using Machine Learning and Monte Carlo Localization https://in5490-run.github.io// Fri, 18 Oct 2019 10:13:12 +0000 Fri, 18 Oct 2019 10:13:12 +0000 Jekyll v3.8.5 Deep CNN-Based Framework For Enhanced Aerial Imagery Registration with Applications to UAV Geolocalization Ahmed Nassar, Karim Amer, Reda ElHakim, Mohamed ElHelw Vegard Bergsvik Øvstegård Fri - 18 Oct 2019 Components Input images Input images On-board camera(Bird’s eye view) Video frame Reference image(Satellite Map) $R$ Known GPS-bounds Not Global Cheating with starting positon $Maths..$ $R$ Calibration Calibration Where is in ? Affine geometric transformation Map $S_i$ with FOV consideration Scale Orientation Remember -> framework is cheating Reduced search space: $r = R(l, w_{s+b})$ Calibrates on $S_1$ and every $S_{(i+3)}$ Scale Invariant Feature Transform(SIFT) Random Sample Consensus(RANSAC) -> Homography matrix Describes movement of UAV related to R. SIFT Homography Sequential frame registration Sequential frame registration Similar to Calibration stage Every non-calibration frame Uses ORB -> Homography -> Translation Oriented FAST and Rotated BRIEF(ORB) ORB Two orders of magnitude faster than SIFT Does not handle scale variance robustly Semantic Segmentation Using U-net U-net U-net Semantic Shape Matching (SSM) Processing U-net frames Fill gaps(Dilation -> Erosion) Remove outlier noise(Erosion - Dilation) Build dictionary of shapes Brute force matching shapes Check matches Calculate homography based on matched shapes Adjust location Semantic Shape Matching (SSM) Morphology Buildings Roads Experiments and Results No real-life testing Only tested on two cities Potsdam Simulated flight with Google Earth 6.3m average drift for local feature mapping 3.6m average drift for SSM Famagusta YouTube Video 10.4m average drift local feature mapping 5.1m average drift for SSM Lower resolution for $R$ and $S$ Used manually-labelled GPS coordinates! Experiments and Results Experiments and Results Experiments and Results Robustness? Computation? Print Back Fri, 18 Oct 2019 00:00:00 +0000 https://in5490-run.github.io//2019/10/18/Vegard/ https://in5490-run.github.io//2019/10/18/Vegard/ presentation Mid-term progress presentation Håkon, Ole og Vegard Mon - 14 Oct 2019 UAV Localization Global vs Local Sensors Inertial navigation system(INS) Visual odometry GPS/GNSS Monte Carlo Localization(MCL) Monte Carlo Localization Problems Robustness: Lighting conditions Environmental changes Optimization: Response time Computational power First priority: Increasing robustness using ML Semantic segmentation Feature extraction Faster local-localization Invariance to Lighting conditions Environmental change 2nd priority: Optimizing MCL Hush-hush / No idea yet System Progress 7% Obtain data Map Data Flight videos Create data-set Test different ML Models U-NET Adaptnet Evaluate models Conduct experiment with video (Optional) Conduct experiment RT on drone Write scientific paper and get world famous Related work Vegard: Deep CNN-Based Framework For Enhanced Aerial Imagery Registration with Applications to UAV Geolocalization Ole Edvin: Vision-based Robot Localization Across Seasons and in Remote Locations, Anirudh Viswanathan, Bernardo R. Pires, and Daniel Huber Håkon: AdapNet: Adaptive Semantic Segmentation in Adverse Environmental Conditions Print Back Wed, 09 Oct 2019 00:00:00 +0000 https://in5490-run.github.io//2019/10/09/Mid-Term-Progress-Presentation/ https://in5490-run.github.io//2019/10/09/Mid-Term-Progress-Presentation/ presentation