It-Jim

Pattern recognition is an extensive field of research since there are a number of things to detect and recognize from the real world. In this project, our goal was to develop an engine for the automatic counting of the various symbols on scanned building plans. Our custom algorithm was able to handle hundreds of different symbols, including complex ones consisting of multiple small sub-patterns and text. It is robust to rotation, scaling and partial symbol occlusions. The developed engine is based on a purely classic computer vision algorithm, works very fast and does not require GPU, which makes it very cheap to use in the cloud. In addition, a comparative test with market competitors indicates an advantage from both accuracy and performance points of view.

KEY POINTS:

1. Automatic recognition of symbols with different complexity
2. Optimized search using multi-threading
3. Custom decomposition algorithm
4. Rotation and scale-invariant recognition engine
5. Successfully integrated into an engineering platform

Case study

The goal of this project was to develop and optimize the image detection and tracking algorithm for AR applications. The main challenge was to make it work directly in the mobile web front-end with all computations done on the edge.
To avoid dependence on internet connection and potential lags, we have built an efficient data processing pipeline for web front-end deployment. Firstly, a custom C++ code was developed for content-based image retrieval and tracking. Secondly, we modified and recompiled the C++ code to the WebAssembly binary code using Emscripten SDK to run it directly in the browser. Finally, the algorithm was fine-tuned and optimized within the Emscripten to ensure proper performance on the web.
Case study

We have developed the positioning algorithm based on cheap Bluetooth beacons and built-in IMU sensors on a mobile device.

BLE signal processing module is based on custom algorithms providing stable positioning results. IMU signal processing module provides robust step detection and device heading estimation on a real-time basis. A specifically developed fusion scheme provides stable and precise user positioning. Our algorithms are implemented in C++ and integrated on Android and iOS platforms.

The positioning engine works in different modes for corridors and open environments like halls. 

Case study
 

Automatic document analysis is a key part of the overall document recognition process. A common scenario is when the user takes a picture by mobile phone or tablet and the goal is to automatically parse and recognize content from the captured document. Such as pictures, tables, text data, links, etc. There are several challenges in this case: geometric distortions of the paper, varying illumination, occlusions. Nevertheless, we have built the page unwrapping engine, which successfully handles the above problems. The developed algorithm contains several key components: Preprocessor, Feature Extractor, Geometric Model Estimator and Refiner. The preprocessor performs image filtering, roughly locates the document boundaries and extracts vertical and horizontal spans (lines and text regions). Feature Extractor module performs parsing of document content. For robustness, we have fused several types of features including corners and edges from image preprocessing in different ways. The output nonlinear grid of points is used as an input for the next system component: Geometric Model Estimator. In order to properly account for the nonlinear geometric distortions of the document, we have constructed a specific 2D-3D model, where the page shape is reconstructed as a 3D surface together with the 6 DoF camera position in 3D space. Finally, after estimation of model parameters, the document is dewarped and most of the existing distortions are corrected. Developed dewarping engine works very accurately and performs distortion correction within a reasonable time frame.

KEY POINTS:

1. Handling illumination changes
2. Correction of nonlinear geometric distortions
3. Automatic page border detection
4. 2-step procedure returns well-refined document layout