基于深度卷积神经网络的行人检测实现

IMPLEMENTATION OF PEDESTRIAN DETECTION BASED ON DEEP CONVOLUTION NEURAL NETWORK

行人检测在视频监控等应用领域具有重要价值。在应用场景复杂、响应速度快的视频监控应用领域,如何提高行人检测的准确率和检测速度是计算机视觉研究者们研究热点之一。深度学习在计算机视觉领域不断创造佳绩,使得深度卷积神经网络在智能监控的通用目标检测中被广泛使用。论文主要介绍一种基于卷积神经网络的行人检测实现方法。该方法以Tensor flow作为训练框架,以Yolo v3作为神经网络算法,使用VOC2007数据集对模型进行训练实现图片的行人检测。试验证明,该方法训练的行人检测模型无论在检测准确度和检测速度,还是模型适用场景,比传统模式的行人检测都有着绝对的优势。

Pedestrian detection has important value in video surveillance and other applications.In the field of video surveillance,the application scenarios are complex and the response speed is fast.How to improve the accuracy and speed of pedestrian detection is one of the hot topics of researchers.With the continuous development of deep learning in the field of computer vision,deep learning convolutional neural network is widely used in the general target detection of intelligent monitoring.This paper mainly introduces the design and implementation of a pedestrian detection system based on convolutional neural network.The system uses tensor flow as the training framework and Yolo V3 as the neural network algorithm.Through the training of the model,the pedestrian detection in the picture can be realized.The experimental results show that the pedestrian detection model trained by this method has an absolute advantage over the traditional pedestrian detection mode in terms of detection accuracy,detection speed and applicable scene.