It-Jim

basketball pitch. The region-of-interest (ROI) control algorithm required pitch mask segmentation for avoiding false detections of the player blobs, panorama stitching from multiple cameras in real time, adaptive filtering and thresholding to detect the player candidates and custom tracker for robust keeping the traces of players. A specific clustering procedure was developed in order to build a proper logic for selection and smooth control of the ROI.

The key advantage of the created tracker is extremely high performance. It works at 120 FPS and multiple system instances can be simultaneously installed in the cloud. This makes the solution quite cheap and flexible, ROI tracking and streaming of the outputs are performed for many clients all over the globe on a real-time basis.

KEY POINTS:

• Multiple objects are tracked on a real-time basis
• The tracker can combine results from multiple cameras
• The solution provides data for sport analytics
• Region of interest for live broadcasting can be extracted automatically

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 images 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 page shape is reconstructed as a 3D surface together with the 6 DoF camera position in 3D space.Finally, after the 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:

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

is robust to rotation, scaling and partial symbol occlusions. Develop 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:

• Automatic recognition of symbols with different complexity
• Optimized search using multi-threading
• Custom decomposition algorithm
• Rotation and scale-invariant recognition engine
• Successfully integrated into an engineering platform

residual warp using the optimized template matcher. Developed tracker is robust to extremal angles, drift-free and very accurate. Algorithms work at 60 FPS on Android and iOS.

KEY POINTS:

• Custom tracker is resistant to extremal angles and drift
• 60 FPS on mobile device (iOS+Android)
• Fusion of feature-based and template-based trackers provides high robustness
• Successfully used in custom mobile AR SDK
  

structure-from-motion (SfM), point cloud densification, meshing and texturing modules. The resulting pipeline was integrated into the web-service, where the user was able to load the data for reconstruction, analyze the resulting 3d model, view the fine details and save the results for further steps.

KEY POINTS:

• 3D point cloud reconstruction engine is implemented
• Successfully integrated into web-service
• Full pipeline including SfM, meshing and texturing is created

without internet connection on mobile devices or tablets. In our project, we have successfully developed a mobile image retrieval system for augmented reality applications. For this purpose, we have optimized and customized the bag-of-words (BoW) algorithm using a properly constructed hierarchical tree for the features vocabulary. As a result, local features from the test images were matched extremely fast. Our CBIR system worked on a mobile device without an internet connection and returned the visual search results within only 2 milliseconds.

KEY POINTS:

• Instant retrieval offline on a mobile device
• 2 ms time for 1000 images database
• Works on iOS and Android
• Integrated into custom mobile AR SDK