17.08.2026 aktualisiert
100 % verfügbar
Computer Vision Engineer | Perception, 3D Reconstruction & Geometry Processing Specialist
Bendorf, Deutschland
Weltweit
Masters in Embedded Systems, Technical University of ChemnitzÜber mich
Computer Vision & ML Engineer mit Erfahrung in 3D Vision, Perception und KI-Daten. Entwicklung von Systemen für Bild-, Video- und Punktwolkendaten sowie Annotation, Geometrieverarbeitung, 3D-Rekonstruktion und ML-Evaluierung.
Skills
Unity3D3D-Computergrafik3D Modellierung3D-Scan3D-RenderingComputer VisionBildverarbeitungObjekterkennungOpenCVComputational Geometry
3D Reconstruction & Geometry Processing
Expertise in NeRF, Gaussian Splatting, Structure-from-Motion (SfM), point cloud processing, multiview geometry, and mesh reconstruction for real-world sensing and visualization applications.
Camera Calibration & Perception Pipelines
Proficiency in camera calibration, multi-camera system orchestration, ICP registration, RANSAC plane fitting, object detection, segmentation, tracking, and end-to-end perception pipeline development.
Programming & Software Infrastructure
Strong command of Python, C++, PyTorch, ROS2, V4L2, Three.js, CMake, GitHub, Docker/Podman, and GCP for building scalable computer vision and robotics software systems.
Simulation & System Tools
Hands-on experience with Panda3D, Unity3D, Blender, MeshLab, NeRF Studio, Meshroom, EasyMocap, ARKit, and RoomPlan for simulation and 3D scene understanding.
Vision & ML Libraries
Practical use of OpenCV, Open3D, Trimesh, SAM/SAM 3D, Vision Transformers, VLMs, and GroundingDINO for computer vision and machine learning tasks.
Annotation & Dataset Tools
Experience with CVAT, LabelMe, Supervisely, Roboflow, CloudCompare, and Segments.ai for dataset creation, annotation workflows, and large-scale data management.
Erfahrung
Berufserfahrung0 - 5 Jahre
Freelancer seit2026
Sprachen
DeutschgutEnglischMuttersprache
Projekthistorie
Developing post-training level-of-detail strategies for large-scale indoor Gaussian scenes without retraining the model. Constructing room-level spatial partitions over trained splats using COLMAP structure and boundary estimation. Implementing renderer-side frustum culling and region-aware activation to reduce active splat count and improve runtime performance on resource-constrained devices.
Built an incremental SfM and multi-camera reconstruction pipeline from scratch. Directed geometry-aware seed-based dense depth propagation from sparse SfM points. Integrated geometric registration and reprojection-based refinement for consistent sparse-to-dense reconstruction. Implemented pose graph construction and reprojection-error minimization for geometry-consistent multi-view reconstruction.
Developed a room-scale 3D geometry processing pipeline using ARKit and RoomPlan, converting depth data into point clouds and meshes through ICP registration, RANSAC plane extraction, and geometric reconstruction. Generated simplified collision meshes and built BVH-accelerated collision models. Trained, optimized, and deployed 3D Gaussian Splat representations for real-time interactive visualization within a Three.js system. Built geometry-aware scene understanding and viewpoint ranking pipelines.