ENGINE MATRIX · 01 — 05
五大自研核心技术体系
Five proprietary core technology engines
Multi-Domain Linked Engine · Data Matrix × Timeline × Filter × Video
01 · Data Matrix × Timeline 多域联动决策引擎
01 · Data Matrix × Timeline Multi-Domain Linked Engine
打破传统软件「看统计是一张静态表、找录像要回时间轴重翻」的割裂感——将交互式交叉矩阵(Data Matrix)、全域时间轴沙盘(Timeline)、多层筛选器与视频队列置于同一内存状态机下毫秒级联动。
Eliminates the divide between static spreadsheets and video timelines—uniting Data Matrix cross-tabs, the Timeline sand table, multi-layer filters, and the clip player in millisecond lockstep.
- 多维交叉点格即播:支持任意原生属性与脚本化组合维度自由交叉(手段×得失、发接×落点、板数×局段),点击任一非空矩阵格即刻生成对应视频连播队列。
- Arbitrary cross-tabs with click-to-play: cross any native or scripted composite dimensions and click any cell to immediately play all matching rallies.
- 筛、表、轴、片四域同源共振:全局筛选条件变更时,Data Matrix 统计值、Timeline 多粒度事件色块高亮区与播放器候选队列同步刷新——筛什么、算什么、轴上见什么、画面播什么严格一致。
- Four-domain synchronization: filter updates simultaneously recalculate matrix cells, highlight Timeline event blocks, and narrow the playback queue on one shared state.
- 指挥沙盘级复盘交互:Timeline 支持 Live 跟赛与 Replay 冻结回看双态切换及帧级边界微调,支撑教练「三分钟定位问题球并当场对上画面」的实战节奏。
- Command-sand-table Timeline: supports Live-follow and Replay-freeze modes plus frame-accurate block trimming so coaches verify patterns in minutes.
≤200ms Capture · Crash-Safe Ring Buffer · CoachCollabCenter ACL
02 · ≤200ms 临场低延迟采集与多席分权协同体系
02 · Low-Latency Capture & Multi-Seat Live Collaboration
让比赛场边升级为「分布式战术雷达站」——主从同场、列级分权、断电防丢、局间秒级可见。
Turns the sideline into a distributed tactical radar—master capture, role-based parallel coding, crash-safe recording, and instant bench visibility.
- 低延迟多源归一化采集:兼容 SDI/USB 采集卡、RTSP/RTMP/HLS 流与屏幕窗口,预览延迟 ≤200ms、打点响应 ≤50ms;采用内存环形缓冲 + 闭合短分段 H.264 写入,意外断电绝不丢失已赛回合。
- Normalized low-latency capture: SDI/USB cards, RTSP/RTMP/HLS streams at ≤200ms preview and ≤50ms tag response, backed by crash-safe segmented H.264 recording.
- CoachCollabCenter 属性列级权限矩阵:Host 主控席管视频与回合起止,多台 Slave 从席按角色分工填报落点、技术手段等专属列,主教练 Viewer 席扫码只读看看板,彻底杜绝多人并发互相覆盖。
- Column-level role ACL: Host controls video and rally boundaries while Slave seats code assigned attribute columns and coaches scan a QR read-only dashboard.
Unified Canonical Fabric · Extensible & Pluggable SportPack / AnalysisPack
03 · 统一技术底座与可扩展、可插拔运动包 / 分析包架构
03 · Unified Technical Foundation & Pluggable SportPack / AnalysisPack
「底层事件骨架高度稳定,上层项目规则与分析口径像插件一样按需热插拔扩展。」
“A rock-solid canonical event core underneath, with hot-swappable sport vocabularies and analytics profiles on top.”
- 统一赛事数据底座(Canonical Event Fabric):全产品线(Suite / Live / WebTag / Analyse / Edit / Play)共享唯一时间锚点与结构化事件模型,显示文案与底层存储键值彻底解耦,一次沉淀全链路复用,杜绝「各模块各算一套数」。
- Unified Canonical Event Fabric: one source of truth across desktop, live, web, analytics, and playback—decoupling UI display labels from underlying storage keys.
- 可插拔 SportPack(运动包)跨项目扩展:将赛制计分规则、球场空间分区、分步动作关系链封装为标准运动包插件,无需重写底层引擎即可平滑扩展乒乓球、网球、排球、羽毛球及篮球/足球等不同球类项目。
- Pluggable SportPack extensibility: encapsulates scoring rules, court zones, and Flow schemas into modular packs across table tennis, tennis, volleyball, badminton, and team ball sports.
- 可扩展版本化 AnalysisPack(分析包):将三段指标、发接轮次、空间热力、失分结构等统计逻辑与呈现模板封装为独立版本化分析包——换运动项目、换教练学派或新增科研口径只需切换/扩展包配置,一键输出 HTML / CSV / 高清 PNG 报告。
- Versioned AnalysisPack plugins: decouples metric formulas from report rendering—adapt to new coaching schools or research KPIs by swapping packs and exporting HTML, CSV, or PNG.
Assisted Event Drafting · Human-in-the-Loop Quality Pipeline
04 · 智能辅助标注与 Human-in-the-Loop 人机协同管线
04 · Assisted Pre-Annotation & Human-in-the-Loop Pipeline
坚持「机器辅助减负、专家复核定音」——适配乒乓球、网球、排球、羽毛球等常见球类项目,让团队从繁重机械切分中解放出来,专注高价值技战术判断。
Built on “machine acceleration, expert authority”—streamlining repetitive rally segmentation across racket and net sports so analysts focus on tactical depth.
- 同构事件草案生成:结合场地几何标定、死球剔除、有效回合起止检测与基础落点/比分初筛,直接输出与人工标注完全同构的时间锚点任务草案。
- Canonical event drafting: combines court calibration, dead-time removal, rally boundary detection, and initial score/zone drafting into standard task structures.
- Human-in-the-Loop 人机回路质控:常规清晰回合快速放行落盘,重度遮挡、擦网等边缘片段自动聚合并提醒人工秒级复核与补标深层战术意图,兼顾提效与竞赛级零容错。
- Human-in-the-Loop review: flags occluded or ambiguous edge rallies for targeted human verification and tactical enrichment, ensuring competition-grade reliability.
FactPack-Constrained Tactical LLM · Clickable Video Provenance
05 · FactPack 约束的可溯源技战术大模型(CoachAI)
05 · FactPack-Grounded Traceable Tactical LLM (CoachAI)
「先由统计引擎算准数字,再由大模型组织洞察,每一句结论都能点回矩阵与原始视频。」
“Compute deterministic facts first, narrate tactical structure second, and link every claim back to matrix cells and video.”
- FactPack 硬约束防幻觉:CoachAI 严禁脱离数据自由猜测比数或得失分率,所有统计数值均来自 Data Matrix 实时计算的确定性结构化事实包。
- Deterministic FactPack grounding: CoachAI never invents statistics; every rate and count is injected directly from verified Data Matrix computations.
- 证据锚点双向穿透 + 离线降级:点击 AI 报告中的任一战术规律或对手习惯,即可联动高亮对应矩阵格并连播原始比赛片段;场馆封闭无外网时自动切换本地确定性规则洞察引擎。
- Clickable clip provenance & offline fallback: click any AI takeaway to highlight its matrix cell and play source clips; degrades gracefully to a deterministic rule engine offline.
多域联动秒看片 · 临场多席协同 · 可插拔底座扩展 · 人机回路与可溯源洞察
Multi-Domain Sync · Live Multi-Seat Collab · Pluggable Packs · Human-in-the-Loop & Traceable LLM