基于区域卷积神经网络的行人检测

Pedestrian Detection based on Region of Convolution Neural Network

行人检测一直是机器视觉领域的研究热点和难点,在智能监控、智能交通和智能机器人等人工智能领域应用越来越广泛。近几年,深度学习尤其是深度卷积神经网络在图像、语音等领域取得了重大突破。针对传统人工设计的特征提取复杂度高且难以有效表达复杂场景中的行人特征问题,提出基于区域卷积神经网络的行人检测算法。该模型通过组合低层特征,形成更加抽象的高层表示属性类别或特征,进而从样本中提取鲁棒性更强、更能刻画图像的特征向量。由于网络模型层次较深,需要训练参数较多,而人工标注行人的数据样本较少,为了防止训练过程中发生过拟合现象,采用微调的方法训练网络。最后,通过多组实验验证,与基于HOG特征的方法相比,该算法能够明显提升行人检测的准确率。

Pedestrian detection is always a hotspot in the field of machine vision, and widely used in the field of artificial intelligence such as intelligent monitoring, intelligent transportation and intelligent robots. In recent years, deep learning, and in particular deep Convolution Neural Network make a significant breakthrough in the image classification, speech recognition and other areas. Artificially-designed methods for feature extracting is hard to implement an imperfect description of pedestrian in the complex background. To solve this problem, a pedestrian detection system based on region of Convolution Neural Network is proposed, adapting a general-purpose convolutional neural network to the task at hand. It can make full use of the advantage of deep convolutional neural network and extract features from the database of pedestrian detection. Because the layer of deep learning network architecture is usually very deep and thus more training parameters are required. The over-fitting problem can be avoided when training the network only on the condition that the training data is sufficient. Finally, several experiments and the comparison with the method based on HOG feature indicate that the proposed our algorithm can obviously improve the accuracy of pedestrian detection.