Abstract:

Object detection is the core of computer vision, which is widely used in image recognition, pedestrian detection, large-scale scene recognition and so on. The emergence of big data and the development of deep learning have injected new impetus into the research of object detection. Traditional target detection mainly uses the method based on manual features and machine learning, which is called feature-based. Present detection algorithm is mainly based on Convolutional Neural Network（CNN）. This paper, firstly analyzes the reason of the poor detection effect of feature-based method and puts forward the improved method. Then, the two-state method and one-state method derived from CNN network are described in detail, at the same time, the connection of each method and the improvement compared with the previous one will be clearly discussed. The principle and process of each network will be described in detail, then the analysis of shortcoming and the effects as well as improvements of the network are given. Finally, the detection effects of different methods on some data sets are included and the existing problems are pointed.

Key words: computer vision, object detection, machine learning, convolutional neural network

摘要：

目标检测是计算机视觉的核心，在图像识别、行人检测、大规模场景识别等方面具有广泛应用，提升目标检测的速度与精度可以拓展计算机视觉的应用范围。大数据的出现以及深度学习的发展为目标检测研究注入了新的动力。传统的目标检测主要使用基于手工特征配合机器学习的方法，即Feature-Based方法。目前的检测算法主要以卷积神经网络（CNN）为核心。分析了Feature-Based方法检测效果差的原因并提出改进方法，详细讨论了CNN网络衍生出的TWO-STATE方法和ONE-STATE方法，介绍了每种方法的联系以及相比之前方法的改进，详细描述了其网络的机理与检测过程，指出了每种方法的检测效果与不足。总结了目标检测方法在一些数据集上的检测效果与仍然存在的问题。

关键词: 计算机视觉, 目标检测, 机器学习, 卷积神经网络