Study And Implementation Of Target Detection System Of Video Based On Real-time Linux System

Target detection technology is an important research content in the fie ld of computer vision and one of the important basic technologies to the problem of understanding image content.Real-time operating system refers to the operating system that can complete the processing of system requests within the specified time.Its main characteristics are timely response and high reliabilit y.Since video-based target detection algorithm has high requirements for computing power and real-time performance,this paper proposes to deploy the target detection algorithm on the real-time operating system and develop a set of target detection system of video based on real-time Linux.And experiments verify that the characteristics of the real-time operating system can improve the real-time performance of the target detection algorithm.The main work of this paper includes:First of all,In view of the ipc hardware system used in this paper,the princ iple and construction process of Xenomai real-time operating system are analyzed,and a scheme of setting up Linux+Xenomai real-time operating system on ipc is proposed.Second,Aiming at the application scene with stable background and single target,the target detection algorithm based on image processing is studied.Based on the frame difference method,three frame difference method and background difference method,an improved algorithm based on the adaptive detection window of target area is proposed.Experiments show that the improved algorithm has better real-time performance than the basic algorithm.Third,Aiming at the application scene with complex background and mult iple targets,the target detection algorithm based on deep learning is studied,and the fully convolut ional neural network in Dlib machine learning library is selected for study and imp lementation.The principle of fully convolutional neural network,the training process and the image processing process when the network propagates forward for target detection are analyzed in detail.Aiming at the hardware and system environment of this paper,an estimation method of computing scale of fully convolutional neural network is proposed,and a method of deploying the network model trained in GPU environment in the real-time operating system environment that bases the target detection system in this paper is also proposed.Fina lly,we find a fully convolut ional neural network model structure suitable for the target detection system in this paper through the experiments.In the end,For the hardware and system environment of this paper,QT is used to develop target detection system of video platform based on real-time Linux,which could integrate the above research contents and facilitate algorithm testing and experimental comparison and verification.Finally,experiments verify that compared with the Linux universal kernel,Xenomai real-time kernel has improved the real-time performance of several target detection algorithms in this paper.