DISTAT CO — Department of Defense STTR Phase II: AF20A-TCSO1

Description

It is difficult to manually count large quantity of small objects fast and accurately for human. Small parts (e.g. screws, bolts, gaskets) have different shapes, sizes, and materials. In this STTR Phase II proposal, we propose an artificial intelligence (AI) software solution to count many small objects that are laid out on a uniform non-reflective surface. By using computer vision (CV) algorithms and machine learning (ML) models, the proposed software system can distinguish, identify and count the objects (laying on a surface with random orientations) based on previously generated machine learning training datasets and real-time images from the mobile device’s camera. In Phase I, we proved the feasibility of training ML models and using mobile app to count groups of small objects in real-time. In Phase II, we are expanding the scope to build a system that can automatically collect and generate unlimited dataset for ML model training and count larger quantity of objects with higher accuracy and speed, using real-time image analysis and process. This system will take into the factors such as object touching each other, overlay on top of each other and clutter together, so that the count will still be accurate even if the parts are not laid out perfectly separated.