Object Detection: Introduction

10.28.04

Have you ever wandered why cars in the 21st century are sill unable to drive by themselves? Or why pilots are still needed to land airplanes? Or why train conductors still exist? One of the reasons is due to the fact that no method has yet been found to detect objects accurately and in a timely fashion. It doesn’t help if the car detects a kid crossing the street 2 minutes later, because by that time the kid will have become road kill.

My project attempts to solve this problem by combining information from two different sensors; an Infra-red sensor (which is very precise but slow in acquiring data) and a set of color images (which is far from accurate). Objects will than be identified from the Infra-red sensor and the images separately and then merged together to more clearly identify objects in a given environment. The goal is that by merging objects from two different sources the precision will be far greater than any current single sensor algorithm.

Formal Definition: Given a distance matrix D (RxC) (taken using the Aibo infra-red sensor) and an image matrix I (R’xC’) taken using the camera from the same location. The matrix D will be incomplete due to the sensor limitations of the infra-red detector. Let the matrix D’ (R’xC’) be a distance matrix but whose granularity corresponds to that of the image.

Problem: Calculate the complete D’ matrix using D and I.