基于异构多处理器和深度学习算法的篮球运动图像目标检测

Object Detection of Basketball Motion Image Based on Heterogeneous Multi-processor and Deep Learning Algorithm

篮球高难度动作识别技术的分析主要是识别和分析篮球运动员在视频中的身体行为.视频识别的目的是提高篮球训练水平.然而,传统的运动目标识别受到场景、动态背景和技术的限制,不能达到预期的效果.因此,本文开发了一种基于深度卷积神经网络的大数据运动目标检测系统,其主要用于篮球运动图像检测.其使用卷积神经网络的高分辨能力来提取图像,以执行计算预处理来识别视频流中的每个人体运动的图像.然后,采用基于Bi-LSTM模型的骨骼识别算法对人体关键点进行检测.最后,开发了一个目标检测系统来重建每个运动.通过选取五组可能导致运动损伤的高难度动作进行实验,结果表明该目标检测系统可以有效提高篮球动作目标识别的准确性,并有助于减少运动员伤病.

The analysis of high-difficulty action recognition technology in basketball is mainly to identify and analyze the physical behavior of basketball players in the video to complete the technical action.The purpose of video recognition is to improve the level of basketball training.However,traditional sports target recognition is limited by the scene,dynamic background and technology,and cannot achieve the desired effect.Therefore,a big data moving object detection system is developed based on deep convolutional neural network for basketball motion image detection,especially to detect basketball motion image.The system uses the high-resolution capability of convolutional neural network to extract images to perform computational preprocessing to identify each human motion image in the video stream.Then,bone recognition algorithm based on LSTM is used to detect key points of human body.Finally,a target detection system is developed to reconstruct each movement.Five groups of difficult actions that may lead to sports injury are selected for experimental study,and the results prove the effectiveness of the proposed target detection system.It is helpful to reduce sports injury.