docs: sync README layout with new backend/frontend modules, add roadmap

- Expand project layout tree with backend/{storage,events,alerting,main}.py
  and the modular frontend (main.js + modules/ + views/)
- Add docs/roadmap.md with the three-phase development direction and link
  it from the README
- List YOLO_DB_PATH in the settings view environment reference

README.md

@@ -13,8 +13,13 @@ src/yolo/defaults.py              Project defaults
 src/yolo/engine.py                Train, validate, predict, and export
 src/yolo/checkpoints.py           Checkpoint discovery and resume support
 src/yolo/reporting.py             Training result reporting
-backend/                          FastAPI inference API and alerting
+backend/main.py                   FastAPI app, detect/events/robots endpoints
-frontend/                         Browser management platform (no build step)
+backend/alerting.py               Consecutive-frame confirmation, WeChat/Feishu delivery
+backend/events.py                 Alert event store (SQLite-backed)
+backend/storage.py                SQLite database, delivery stats, UTC timestamps
+frontend/main.js                  Browser entry, view switching, boot sequence
+frontend/modules/                 Shared helpers (dom, api, ui, charts, state)
+frontend/views/                   Per-view rendering (dashboard, inspection, events, robots)
 ```
 
 ## Setup
@@ -100,3 +105,7 @@ Detection requests without a `session_id` share the fixed `single-image` session
 - `data/Smoke-Fire-Detection-YOLO/` — main smoke/fire dataset (train/val/test).
 - `data/fire-dataset/` — incremental hard-negative collection (395 train / 87 val images, Pascal VOC XML plus YOLO labels); not yet wired to a training config.
 - `runs/audit/` — label audit tooling: `scan_missing_labels.py` finds unlabeled images that the current model detects, `render_missing_label_candidates.py` renders them for review.
+
+## Roadmap
+
+The development direction (hard-negative fine-tuning, YOLO26n comparison, real-time push, multi-video concurrency, data flywheel, and more) is tracked in [`docs/roadmap.md`](docs/roadmap.md).

docs/roadmap.md

@@ -0,0 +1,41 @@
+# 发展路线图
+
+均衡推进：模型能力与平台能力并进，按依赖关系分三阶段。每阶段结束都有可独立验证的增量。
+
+## Phase 1：数据与模型闭环 + 实时化
+
+1. **难负样本微调落地**
+   - 为 `data/fire-dataset`（train 395 / val 87，Pascal VOC XML + YOLO 标签）编写数据集配置 `configs/datasets/smoke_fire_hard_negative.yaml`，VOC XML→YOLO 转换脚本纳入 `src/yolo/data/`
+   - 执行 `TrainConfig` 已预设的运行名 `smoke_fire_yolo11s_hard_negative_ft_v1`，与 v1-4 对比验证误报下降
+2. **YOLO26n 评估补齐**
+   - 对 `runs/detect/smoke_fire_yolo26n_v2-3/weights/best.pt` 跑一次 `fire-yolo val`，补齐 `args.yaml`、`results.csv` 与曲线图
+   - 更新 `docs/model_comparison.md`，完成部署选型对比（nano 对 CPU/边缘更友好）
+3. **SSE 实时推送**
+   - 后端新增 `GET /api/stream`（SSE，事件与机器人状态变更时广播）
+   - 前端 `EventSource` 替代总览/告警中心的轮询刷新
+
+## Phase 2：并发产能 + 安全加固
+
+4. **多路视频并发巡检**
+   - 推理锁升级为单 worker 专用执行器 + 请求队列（当前 FastAPI 线程池在锁排队时可能饿死其他端点）
+   - 前端多路视频格布局，每个视频独立会话
+5. **ONNX / TensorRT 导出与加速**
+   - `fire-yolo export` 已具备，补充部署文档与延迟/吞吐基准对比
+6. **事件截图存档**
+   - 告警触发时截图落盘 `data/screenshots/<event_id>.jpg`（数据库存路径），前端事件详情弹窗展示
+7. **鉴权收紧**
+   - 简单 Bearer token + CORS 白名单；在对外/多用户访问需求出现前完成，避免带病部署
+
+## Phase 3：工程化 + 数据飞轮
+
+8. **audit 集成进 CLI**
+   - `fire-yolo audit` 参数化现有 `runs/audit/scan_missing_labels.py` 与 `render_missing_label_candidates.py`，输出保持 CSV/JSON 格式
+9. **误报数据飞轮**
+   - 前端"误报"按钮 → 截图+检测框入库 → 定期导出为增量难负样本 → 再微调
+   - 依赖 6 与 8，形成"数据 → 模型 → 数据"的复利闭环
+10. **Docker 化**
+    - CUDA 基础镜像 + uv + 卷挂载 `data/`，单机部署稳定后再固化
+
+## 排序逻辑
+
+先"数据/模型闭环"（依赖最少、价值最大）→ 再"实时化"（值班体验增益）→ 后"并发/加速/安全"（相互依赖且依赖前两者）→ 最后"工程化与飞轮"（依赖前面所有组件）。

frontend/index.html

@@ -199,6 +199,7 @@ YOLO_DEVICE=0
 YOLO_IMGSZ=768
 YOLO_CONF=0.40
 YOLO_IOU=0.45
+YOLO_DB_PATH=data/app.db
 ALERT_CONFIRM_FRAMES=3
 ALERT_COOLDOWN_SECONDS=60</code></pre></article>
           </section>