- Developed a multi-camera calibration method and calibrated 70 cameras in a single space, achieving triangulation accuracy of less than 3 cm- Identified and reported errors in the OpenCV module related to 2D marker detection- Proposed a calibration method using LIDAR, which provided similar accuracy but was easier to use compared to the method using only cameras- Created a real-time algorithm for human pose estimation, achieving accuracy within 5 cm despite low detection precision and unsynchronized cameras

- Laid a solid foundation for the successful development of software for optical navigation of UAVs- Coordinated and planned technical workflows for UAV optical navigation software development- Created software architecture for SLAM, obstacle avoidance, drone control, path planning, and sensor integration- Developed integration testing and benchmarking approaches- Set up and integrated CI and CD into the development process

- Optimized linear algebra algorithms, achieving performance improvements ranging from 10x to 1000x - Maintained the [Metric](https://github.com/metric-space-ai/metric) library: created unit tests and benchmarks with Catch2, set up pre-commit hooks and CI, modernized the build system with CMake, and revised the documentation- Implemented statistical algorithms- Created a tiny DNN engine with dense and convolutional layers from scratch- Optimized the Daubechies wavelets algorithm using the Blaze linear algebra library to enhance performance and efficiency- Created a build system for a python project with embedded C++ code using modern CMake, Conan, pybind11, and scikit-build- Created a Conan [package](https://conan.io/center/recipes/wavelet_buffer) for WaveletBuffer and added it to conan-center-index

- Led the development of a SLAM backend engine from scratch, completing the project within three months. Efficiently organized team workflow, achieving 20% higher accuracy compared to a solution developed by another department over the course of a year- Developed and implemented a real-time algorithm for precise camera localization using 2D barcodes, achieving 10 times the accuracy of standard methods- Developed an original Python algorithm for determining the spatial position of a VR controller with a dot pattern, achieving performance several times faster than other teams’ solutions- Conducted research and testing of GTSAM as part of a review of modern SLAM solutions, identified a critical issue that made it less effective than Ceres Solver for SLAM applications, and communicated the problem to the author on GitHub, resulting in its resolution issue)- Researched methods for representing urban landscapes in SLAM as planes and geometric primitives, aiming to reduce memory usage and improve landscape part matching

Completed many different projects, providing a broad view of the industry. Handled all stages independently, from the creation of the terms of reference to development, communication, and project delivery. This experience provided a comprehensive understanding of the entire development cycle and related areas, beyond just programming- Achieved 100% Job Success Rate, earned Top Rated Status, and received an 85% Client Recommendation Rate on Upwork- Designed and implemented a stereocamera system with two single cameras, including calibration and synchronization- Developed an application for one-hand large gesture recognition using a single-lens camera- Created a sportsmen ID identification system using a self-developed 2D-barcode with checksum- Engineered a hockey training device with two high-speed cameras to detect puck trajectory, velocity, and hit points- Developed a UAV landing system tracking a QR code with a single camera, achieving high accuracy in poor light and weather conditions- Built a video search engine to find modified clips (color, crop, reflection) within a video database- Processed drone aerial photos to produce orthophoto maps and 3D reconstruction models

- Ported mobile application from Android to Samsung Bada OS

- Developed a custom augmented reality engine from scratch in C++, optimized for real-time performance and low hardware requirements- Implemented hand gesture recognition using Kinect to create an interactive user interface- Created a face tracking application capable of detecting, cropping, morphing, and overlaying faces in video segments automatically- Led a team of two engineers, overseeing project planning and execution