Object Detection Using Deep Convolutional Neural Networks

Abstract

Object detection is a fundamental problem in image analysis and understanding. Lots of progress have been acquired on object detection due to the introduction of deep convolutional neural networks in recent years. Most of those algorithms can be categorized into two types, the two-stage method composed of region proposals generation and object classification along with the position regression of bounding box, and the one-stage regression method directly predicting classes and anchor offsets of objects. In this paper, the typically explored methods of these two types will be discussed to illustrate their development procedures. Moreover, Faster R-CNN and SSD are chosen as representatives for comparison. Experimental results demonstrate that the detection accuracies of SSD and Faster R-CNN are close, and each has its own merits in different images.