Introduction

This project came from an idea to locate movement in the real world for a toy turret to target. The idea then evolved into learning the features of an object and tracking it in real time with a camera, without the toy turret.

Problem Statement

Current camera tracking requires cameras with a large field of view, a human operator, or a primitive form of object tracking. There are few solutions that have attempted to solve the problem of tracking specific objects or people in real time.

Potential Applications

• Military

• Drones: Following a person/object around autonomously

• Security Cameras

Goals

Trakam is a combination of a typical panning security camera, machine vision and machine learning. The final product will ideally implement a model for quickly identifying an object that enters the camera’s field of vision and successfully maintains the object in the field of vision as long as possible. Goals are as follows in chronological order of completion:

• To use machine vision and machine learning to detect objects within its field of vision.

• To keep a high priority object within the field of vision (starting with one). Switching between objects as higher priority objects enter the field of view.

• To build a client that is able to view what the unit is seeing. The client should also be able to set tracking targets and change priority of current targets.

• To be able to differentiate between individual humans.

• To keep multiple high priority objects in the field of view trying to center the highest priority object.

• To be able to use multiple Trakam units to follow objects as they move out of the field of view of an individual unit.

Parameters

The members are cognizant of the difficulty of extremely high precision with such a task. As a result the primary parameters of the project are the following:

• Track with low latency

• Maintain relative cost efficiency

• Smooth (minimal jitter) video feed