67 lines
1.8 KiB
Markdown
67 lines
1.8 KiB
Markdown
# RK3588 安装与验证
|
||
|
||
## 环境
|
||
|
||
- AArch64 RK3588
|
||
- Ubuntu 22.04 / glibc >= 2.35
|
||
- 可用的 RKNN NPU 驱动
|
||
- 不需要安装系统 OpenCV；标准 SONAME 的 OpenCV 4.5.0 和 `librknnrt.so` 已包含在 `lib/`
|
||
- 不需要 AI Studio Runtime
|
||
|
||
## 解包
|
||
|
||
```bash
|
||
tar xzf stereo_bolt_delivery_mainline_20260714.tar.gz
|
||
cd stereo_bolt_delivery_mainline_20260714
|
||
export LD_LIBRARY_PATH="$(pwd)/lib:${LD_LIBRARY_PATH:-}"
|
||
chmod +x validation/c_example_centerline \
|
||
validation/selected4_delivery_validate \
|
||
validation/run_selected4.sh
|
||
```
|
||
|
||
## 完整性与依赖
|
||
|
||
```bash
|
||
sha256sum -c MANIFEST_SHA256.txt
|
||
ldd lib/libstereo_bolt.so | grep "not found"
|
||
# 第二条命令预期无输出
|
||
```
|
||
|
||
关键文件预期 SHA256：
|
||
|
||
```text
|
||
5e7884a9f68ba389105d2af17f57482f4a15c2953f7845ead636653ae2d0aebd lib/libstereo_bolt.so
|
||
b9ed3b46c416919dc969637a7093cf01312eda91320066e40e33daad56d9f169 weights/model_20260713T084621+0000_c732327c.rknn
|
||
d9157898f966927b38cbe4d9531baa9ea514c28153001654d1e2d5c2c4ad14de config/config.rk3588.mainline.yaml
|
||
```
|
||
|
||
## 四帧板端复验
|
||
|
||
```bash
|
||
bash validation/run_selected4.sh
|
||
```
|
||
|
||
脚本会分别执行 YAML 文件接口和显式模型 buffer 接口，并检查 19、20、28、43 四帧。每帧应包含：
|
||
|
||
```text
|
||
PRECISE ... success=1 yolo_left=8 yolo_right=8 pairs=8 bolts=8
|
||
RANGE ... found=1 bolts=8 status=ok
|
||
VERDICT=PASS
|
||
```
|
||
|
||
最终必须出现：
|
||
|
||
```text
|
||
=== PACKAGE_REVERIFY_PASS ===
|
||
```
|
||
|
||
## 原接口调用
|
||
|
||
文件入口保持不变：
|
||
|
||
```c
|
||
StereoBoltCtx* ctx = sb_create("config/config.rk3588.mainline.yaml");
|
||
```
|
||
|
||
生产 buffer 入口仍使用 `sb_create_ex()`、`sb_process_bolt_module_buffers()` 和 `sb_range_bolt_binned()`；调用方只需继续包含 `include/stereo_bolt/c_api.h`。配置中的模型和标定路径均相对于解包根目录解析。
|