🔐 {{lang==='zh'?'登录后可体验跌倒检测模拟器':'Login to experience the fall detection simulator'}}

{{lang==='zh'?'系统结构悬浮拆解 · 核心部件解析':'System Architecture Exploded View · Core Component Analysis'}}

📊 {{lang==='zh'?'单套硬件成本拆解':'Unit Hardware Cost Breakdown'}}

{{lang==='zh'?'本系统以Milk-V Duo S（SG2000）国产RISC-V开发板为主控核心，搭载SG2000双核异构芯片（1GHz C906 RISC-V核+1GHz A53核），512MB DDR3内存，0.5TOPS INT8独立TPU。外扩OV2640 USB摄像头、MPU6050六轴陀螺仪、DHT11/MQ-2环境传感器、SPI SD卡模块，构建视觉+惯性双模态居家康养监护硬件平台。自研四层扩展板解决原生开发板外设供电不稳问题，全部器件为通用开源模块。':'This system uses Milk-V Duo S (SG2000) domestic RISC-V development board as the main controller, featuring SG2000 dual-core heterogeneous chip (1GHz C906 RISC-V + 1GHz A53), 512MB DDR3, 0.5TOPS INT8 dedicated TPU. Expanded with OV2640 USB camera, MPU6050 6-axis gyroscope, DHT11/MQ-2 environmental sensors, SPI SD card module — building a visual+inertial dual-modal home care hardware platform. Custom 4-layer expansion board solves native board peripheral power instability; all components are standard open-source modules.'}}

🧩 {{lang==='zh'?'主控芯片 · Milk-V Duo S (SG2000)':'Main Controller · Milk-V Duo S (SG2000)'}}

📐 {{lang==='zh'?'惯性传感单元 · MPU6050':'Inertial Sensing Unit · MPU6050'}}

📌 {{lang==='zh'?'核心引脚资源分配':'Core Pin Resource Allocation'}}

🖥️ {{lang==='zh'?'硬件拓扑框图':'Hardware Topology Diagram'}}

Milk-V Duo S (SG2000) {{lang==='zh'?'双核异构':'Dual-Core Heterogeneous'}}
├── RISC-V C906 @1GHz → {{lang==='zh'?'系统调度+传感器融合':'System Scheduling + Sensor Fusion'}}
├── ARM A53 @1GHz → AI {{lang==='zh'?'推理 (可选)':'Inference (Optional)'}}
├── TPU 0.5TOPS INT8 → YOLOv8-Pose {{lang==='zh'?'姿态模型推理':'Pose Model Inference'}}
├── I²C1 → MPU6050 {{lang==='zh'?'六轴陀螺仪':'6-Axis Gyroscope'}} (100Hz)
├── GPIO → DHT11 + MQ-2 {{lang==='zh'?'环境传感':'Environmental Sensing'}} (1Hz)
├── SPI3 → SD{{lang==='zh'?'卡':' Card'}} AES-128 {{lang==='zh'?'加密存储':'Encrypted Storage'}}
├── USB → OV2640 {{lang==='zh'?'摄像头':'Camera'}} (640×480 22FPS)
├── WiFi6 → MQTT{{lang==='zh'?'远程脱敏告警推送':' Remote Sanitized Alert Push'}}
└── {{lang==='zh'?'自研四层扩展板':'Custom 4-Layer Expansion Board'}} → {{lang==='zh'?'稳压+浪涌保护+总线引出':'Regulation+Surge Protection+Bus Breakout'}}

{{lang==='zh'?'本算法针对居家独居老人跌倒检测难题，提出"视觉初筛 → 惯性校验 → 时序回溯"双模态融合判定框架。区别于传统单一视觉阈值检测方案，双模态设计有效解决了弯腰、下蹲等居家相似行为与真实跌倒的信号混淆问题，将误报率压制至0.7%以下，暗光遮挡场景漏报率降至0%，同时保持召回率>95%。':'This algorithm addresses the challenge of fall detection for elderly living alone, proposing a "visual pre-screening → inertial verification → temporal backtracking" dual-modal fusion judgment framework. Unlike traditional single visual threshold detection, the dual-modal design effectively resolves signal confusion between similar home behaviors (bending, squatting) and real falls, suppressing false alarm rate to below 0.7%, reducing low-light/occlusion miss rate to 0%, while maintaining recall >95%.'}}

🔁 {{lang==='zh'?'双模态算法决策流程':'Dual-Modal Algorithm Decision Flow'}}

{{lang==='zh'?'摄像头帧读取(22FPS) → YOLOv8-Pose TPU推理(45ms) → 17关键点姿态初筛':'Camera Frame Capture(22FPS) → YOLOv8-Pose TPU Inference(45ms) → 17-Keypoint Pose Pre-Screening'}}
    ↓ {{lang==='zh'?'初筛触发':'Pre-Screen Trigger'}}
MPU6050{{lang==='zh'?'惯性校验(俯仰角>28° & 合加速度>2.2g) → 双条件同时满足':' Inertial Verification(Pitch>28° & SVM Accel>2.2g) → Both Conditions Met'}}
    ↓ {{lang==='zh'?'双模态触发':'Dual-Modal Trigger'}}
8{{lang==='zh'?'帧时序回溯过滤 → 排除弯腰/下蹲误动作 → 确认跌倒 → 告警触发':' Frame Temporal Backtrack Filtering → Exclude Bending/Squatting False Actions → Confirm Fall → Alert Trigger'}}

⚙️ {{lang==='zh'?'可调阈值参数表':'Adjustable Threshold Parameters'}}

{{lang==='zh'?'本软件框架基于裁剪版Buildroot Linux系统（内存占用从280MB压缩至92MB），采用C++编写底层外设驱动与多传感器数据融合逻辑，ncnn轻量化推理框架运行INT8量化YOLOv8-Pose模型。全部代码基于MIT协议开源，模块化设计便于二次开发与高校RISC-V实训教学。':'This software framework is based on a trimmed Buildroot Linux system (memory footprint reduced from 280MB to 92MB), using C++ for low-level peripheral drivers and multi-sensor data fusion logic, with ncnn lightweight inference framework running INT8 quantized YOLOv8-Pose model. All code is open-sourced under MIT license, with modular design facilitating secondary development and RISC-V training in universities.'}}

🔍 {{lang==='zh'?'田野调研':'Field Research'}}

📍 {{lang==='zh'?'调研地点：十堰市郧阳区白桑关村、柳陂村、青曲村':'Research Locations: Baisangguan Village, Liubei Village, Qingqu Village, Yunyang District, Shiyan City'}}

📈 {{lang==='zh'?'技术性能验证':'Technical Performance Verification'}}

🏘️ {{lang==='zh'?'试点成果与外部支持':'Pilot Results & External Support'}}

📅 {{lang==='zh'?'项目进度里程碑':'Project Milestones'}}

{{lang==='zh'?'团队覆盖嵌入式系统、AI算法、硬件设计、系统集成四个专业方向，形成"感知—算法—系统"完整技术链条。成员具备RISC-V嵌入式开发、传感器应用、AI模型部署、PCB硬件设计等跨学科技能，在多项国家级创新创业竞赛中取得优异成绩。':'The team covers four professional areas — embedded systems, AI algorithms, hardware design, and system integration — forming a complete "sensing-algorithm-system" technical chain. Members possess cross-disciplinary skills including RISC-V embedded development, sensor applications, AI model deployment, and PCB hardware design, with outstanding achievements in multiple national innovation competitions.'}}

🎓 {{lang==='zh'?'指导教师':'Advisors'}}

{{lang==='zh'?'刘国阳':'Liu Guoyang'}} ({{lang==='zh'?'智能网联汽车学院':'School of Intelligent Connected Vehicles'}})   {{lang==='zh'?'王凯旋':'Wang Kaixuan'}} ({{lang==='zh'?'创新创业教育学院':'School of Innovation & Entrepreneurship Education'}})

📋 {{lang==='zh'?'项目任务分工表':'Project Task Assignment'}}

{{lang==='zh'?'基于项目BOM成本核算（单套硬件综合成本542元），针对不同客户群体制定差异化定价策略。批量采购可显著摊薄固定成本，实现更低单价。整机物料成本仅为商用同类产品的1/3，且无后续云服务年费。':'Based on project BOM cost accounting (unit hardware cost ¥542), differentiated pricing strategies are formulated for different customer groups. Bulk procurement significantly reduces fixed costs for lower unit prices. Total BOM cost is only 1/3 of commercial counterparts, with no recurring cloud service fees.'}}

📊 {{lang==='zh'?'单套硬件成本拆解':'Unit Hardware Cost Breakdown'}}

💲 {{lang==='zh'?'差异化定价方案':'Differentiated Pricing Plans'}}

📋 {{lang==='zh'?'大创项目经费预算':'Innovation Project Budget'}}

{{lang==='zh'?'模拟MPU6050陀螺仪数据+视觉姿态初筛结果，演示双模态跌倒检测判定逻辑。':'Simulate MPU6050 gyroscope data + visual pose pre-screening results, demonstrating dual-modal fall detection judgment logic.'}}