Third eye

 Our objective is to enhance military and search & rescue operations by quickly identifying people that are behind obstacles or under rubble. In both of these critical applications where time is of the essence this project shows significant value in reducing search times and allowing people to be rescued quicker. 

The proof of concept uses an Intel RealSense depth camera and the MoveNet pose estimation model to identify and localize people in its frame. The location of the people identified by the depth camera is then relayed to a second monocular camera which cannot see the people. Then the location of the identified people are highlighted on the screen of the second camera to make them visible. 

The system showed promising results in this simple test environment. While the location of the person was accurately identified most of the time, the results became less accurate when the subject was more than 8 meters away from the camera. This indicates a problem as the plan is to put this technology on to drones. To overcome this distance limitation, the system should be tested with other methods that can identify the location of the person such as using a lidar sensor or radar.