Adaboost Cascade Classifier for Classification and Identification of Wild Animals using Movidius Neural Compute Stick

Abstract

Deep learning has gone deeper and has been utilized in almost all the applications like object recognition, image classification, speech recognition and much more. Most of the real-time applications rely on deep learning for accurate results. But one downside to the deep learning is its demand for GPUs (Graphical Processing Unit) or TPUs (Tensor Processing Units) for faster execution. There was no one-stop-shop hardware and software for deep learning applications, until recently Intel launched the Movidius Neural Compute Stick (NCS). This sleek device provides the power of GPU in a CPU based system. In this work, we have modeled an animal detection system using NCS and AdaBoost classifier powered by Multi-Block Local Binary Pattern (MB-LBP) features. The model has been built upon AlexNet and has achieved an average accuracy of 96.8% and a false rate 2.3% in classifying the animals as wild and non-wild. Furthermore, the model has a speedup of 467% when compared to the execution in the CPU based system.