Non Rule based Artificial Intelligence
- FACIAL EXPRESSION : A recognition system built on publicly available data set plus Faces created using Generative Adversarial Networks which involves face detection (green square), facial landmark detection (blue crosses), registration to reference landmarks (red circles), and illumination correction. Currently the engine is identifying 6 expressions (neutral, surprised, sad, happy, fear, angry)
- GESTURE ANALYSIS: Instead of going with traditional gesture object detection, our approach is based on finding new patterns from ‘body part’ estimations which will rule out the dataset requirement and is undefined so newer non-trained pattern can also be recognized.
- AUDIO ANOMALY: System can also be pattern trained on short audio patterns that will identify the audio pattern and be passed to text2speech module which in turn will produce justifiable actionable action keywords (e.g. fight, normal gathering, protest, etc.). Combining this with above other 2 analytics will produce a nearly accurate forensic analysis of the scene.
