Abstract
The V1KU is a product by Cognimem Technologies, Inc., which combines a hardware neural network chip, a Micron/Aptina monochrome CMOS sensor (camera) and CogniSight image recognition engine. It is capable of learning the target either by giving it examples of the object to track using pre-captured images or by using the included camera. This project extends the functionality of the V1KU module for the purposes of tracking. Two V1KU modules are inserted into a camera mounting system that allows the V1KU modules to tilt vertically similar to the way the human eyes move coordinated up and down. The harness also allows for horizontal movement of each module individually. In this configuration, the V1KU modules are able to stereoscopically track an object in all 3 dimensions. The application combines the facilities to teach the V1KU modules a given object and track that object. The program calculates the position of the identified object using the pixel coordinates and servo angles. It then uses that information to keep the target in the center of the camera for each module. A Kalman filter-tracking algorithm is to predict the next location of the object in case the tracked object becomes obstructed or un-identified to a short period. The result is a tracking solution that can follow any learned target seen using its cameras alone.