Discrimination of Birds from Aircraft in a Video Sequence for EO Tracking Systems Using Deep CNN

Birds flying in front of cameras of Electro Optical tracking system distracts video tracker from tracking flying objects like aircrafts. This intrusion always challenges the operators during automatic tracking of aircrafts in test ranges. With the advent of CNN and consequently deep neural network, object classification became more accurate and robust. In this paper we have fine-tuned and configured one deep neural network to classify birds and aircraft in video sequences which is used as supplementary cue for conventional video tracker. A seeded region growing segmentation followed by feature based tracking using Kalman filter is used to localize the target position correctly. This is followed by a modified AlexNet for classification alongside a FFT based flapping motion detector. This method provides a very accurate classification which can aid the tracker not to lose track during an aircraft auto track. The experimental result with the curated dataset shows promising result in terms of speed and accuracy.