Deployment of Object Detection Enhanced with Multi-label Multi-classification on Edge Device

This paper presents a method that enables detecting multiple mutually exclusive features of each class using the same convolutional neural network (CNN). This method proposes the design of a Multi-Label Multi-Classification (MLMC) deep learning model on resource-constrained edge devices. The proposed technique is also extended to utilizing the same CNN for classifying heterogeneous datasets. The MLMC model is used in conjunction with object detection techniques that leverage on a cropping system. The combined model used in conjunction with Intel's Neural Compute Stick (NCS) achieves the detection of an object of interest at the rate of 10.51FPS out-performing conventional object detection speed on edge devices.