老人摔倒识别设计报告.pptx

老人摔倒识别西安电子科技大学团队名称：RandomVariable小组成员：赵鑫玮吕卡夫冯开元指导老师：李隐峰华为指定选题B6

目录CONTENTSRandomVariable项目背景01系统框架03算法优化02作品展示04

BACKGROUND项目背景01老人摔倒

ALGORITHM算法优化02RandomVariable

跌倒问题倒不倒100/1分类器Logistic回归？SVM?神经网络因素1：准确率RandomVariable因素2：训练时间

RandomVariable

RandomVariable90/0前摔后摔左摔右摔pitch180000roll00-9090X~~↗↘Y↗↘↗↗Z↗↘~~90/0前摔后摔左摔右摔pitch018000roll0090-90X~~↘↗Y↘↗↗↗Z↘↗~~

90/0前摔后摔左摔右摔pitch180000roll009090X~~↘↗Y↘↘↘↘Z↗↘~~90/0前摔后摔左摔右摔pitch018000roll009090X~~↗↘Y↘↘↘↘Z↘↗~~RandomVariable

90/0前摔后摔左摔右摔pitch180000roll00-9090X~~↗↘Y↗↘↗↗Z↗↘~~90/0前摔后摔左摔右摔pitch180000roll009090X~~↘↗Y↘↘↘↘Z↗↘~~RandomVariable

RandomVariable1.2S完成采样率80Hz序列长度1016个属性，共计606个输入（特征）1个输出→0/1?1.2S完成采样率80Hz序列长度101

RandomVariable?

RandomVariable可行性分析左摔序列样本前摔序列样本自由跌落序列样本走路序列样本

RandomVariable数据采集

RandomVariable正则化系数00.10.20.30.40.511.52神经元个数24训练集准确率96.6497.4498.1297.9897.7197.8595.6992.1990.58检验集I准确率89.3890.0090.0091.8890.0091.2587.5083.1381.25检验集II准确率89.3191.8289.9491.1989.3189.9490.5786.7983.6525训练集准确率97.9898.3898.3897.7197.7197.8593.9492.0691.12检验集I准确率86.8891.2591.2590.6389.3891.2585.6382.5081.88检验集II准确率90.5790.5793.0888.6889.3191.8286.7986.1684.9126训练集准确率97.3197.9897.9897.7197.7197.8595.6992.1990.58检验集I准确率89.3890.6391.2589.3889.3891.2587.5083.1381.25检验集II准确率89.9488.6891.8289.9488.6889.9490.5786.7983.65

RandomVariable不存在过拟合,欠拟合问题

RandomVariable门限0.30.350.40.450.50.550.60.650.70.75漏报率0.38%0.38%0.47%0.66%0.66%0.85%1.41%1.88%2.54%3.86%误报率3.48%3.01%2.35%1.98%1.69%1.51%1.32%1.04%0.94%0.75%真实摔倒真实未摔倒预测摔倒真阳性误报预测未摔倒漏报真阴性??门限0.30.350.40.450.50.550.60.650.70.75漏报率0.38%0.38%0.47%0.66%0.66%0.85%1.41%1.88%2.54%3.86%误报率3.48%3.01%2.35%1.98%1.69%1.51%1.32%1.04%0.94%0.75%

系统框架03设计思路及系统框架HOWTOCONDUCT

18Forward团队系统框架RandomVariable

04PRESENTATION作品展示ImagineCup2014成果汇报、效果展示

RandomVariable作品展示

RandomVariable作品展示

RandomVariable作品展示

RandomVariable实验结果?前摔侧摔后摔合计测试组数505050150正确识别组数494743139漏报率2%6%14%7.33%?走路慢跑摔落放口袋合计测试组