ANALYTICAL SOLUTIONS

We develop machine learning based computer vision algorithm for identifying armed individuals using visible and infrared cameras. We compile and develop custom datasets to train and benchmark our vision algorithms to accurately and efficiently detect threats in urban and natural settings. Images sourced from the public domain are annotated manually. We also create synthetically generated images which are auto annotated saving manual labor for data preprocessing. Synthetic training datasets are created using Poisson-blending and Generative Adversarial Networks for in a variety of different scene conditions for higher classification accuracy. We also explore virtual 3D environments rendered by game engines such as Unity to create more realistic synthetic training data. A 3D simulated environment allows the possibility capturing the object in different camera angles, lighting conditions, and background noise and provides a sandbox approach to artificial intelligence and ML... We are..