I did lot of research as well developed this software system using various Machine learning methods. I have spent around one year on this project to implement this technology for a local state government. Unfortunately It didn't materialised. But I am interested in contributing to open source community. It can accurately identify, segment, recognise objects in video feeds (92 types of semantic attributes of a person in video feeds). The most interesting part is the accuracy of our facial recognition of wild shots from street cctv cameras.
EagleEye is a surveillance system based on true Artificial Intelligence. This digital brain evolved over time to recognize face biometrics, people behaviors and objects even in difficult outdoor environments. It is the most versatile, accurate, faster video surveillance technology.
Real time crime prevention
As system process video streams in real time, certain events are predicted and sent to intelligent alarm such as face match, license plate match, suspicious objects, person attributes match. With the help of intelligent alarm, we can track and find target person as well as prevent potential crime in future
Post crime investigation
There are situations in which we need to find specific person or vehicle for post investigations. Post records search enables us to meet this requirement. We can search suspect or missing person face image, car number, his/her attributes in our database. We can even filter those results to short list findings. Thus, enables an efficient investigating tool for policing.
Real time predictions
System process thousands of video streams in real time. It constantly generates predictions for all pipelines – Face, vehicle plate, semantic attributes.Whole system is designed to scale and for predicting in Realtime.Whole system is carefully designed to run all operations simultaneously. System is adapted to process high of amount visual data and predict faces, behaviors and objects in the Realtime which means One step ahead to prevent potential crimes in future.
Adapted for outdoor surveillance
Outdoor conditions are very different for digital image capturing compared to studio quality face shots. We solved this problem by training custom neural networks to predict accurately in such environments.
- Different temperatures effects huge change color
- Rainy, fog effects cause blur
- Low fps cameras cause motion blur
- Unpredictable viewing angles difficult to obtain facial and license plate data
I am very much interested to continue this project and make it open source if people are also interested.