• Currently working as a senior machine learning engineer at Calipsa.
  
  
• Calipsa’s AI platform analyses CCTV alarms in real-time to filter out false ones and automatically notify alarm monitoring services about true alarms.
  
  
• My focus at Calipsa is to develop ML models that detect motion of people and cars in videos (true alarm), as opposed to motion triggered by changes in light conditions or weather, by animals or other events that are not in our target (false alarm).
  
  
• Challenges include being able to process very short videos (usually 2-3 video frames), settings with low light levels or low contrasts, detecting small amount of motion or small cars or people, detecting occluded people or cars, and unseen settings (generalisation).
  
  
• Currently, the ML models achieves an average of 94% reduction while maintaining a recall of 99+%.
  
  
• I also research methods for transfering models to low resolution (as low as 60px x 80px) settings and thermal settings where data is currently less readily available (5000 training data points for these settings).