A Performance Comparison between Deep Learning Network and Haar Cascade on an IoT Device

This paper seeks to present the development, operation, and comparison of two object recognition methods trained for the classification of different objects ranging from simple geometric forms to traffic signs or car registration plates. There are many techniques, which have been proposed for object detection but this paper focuses on two of them: Haar Cascade Classifier (a classic method for face detection) and Deep Neural Network (a modern approach). Furthermore, different methodologies and training methods are tested on different devices including an IoT device and comparing both performances. In the end, the authors propose several accuracy enhancements of one of the methods using another one, and the drawbacks between methods on low-power device.