分析时间戳

SDK 提供了时间戳分析的脚本 stamp_analytics.py 。工具详情可见 tools/README.md 。

参考运行命令及结果，于 Linux 上：

$ python tools/analytics/stamp_analytics.py -i dataset -c tools/config/mynteye/mynteye_config.yaml
stamp analytics ...
  input: dataset
  outdir: dataset
open dataset ...
save to binary files ...
  binimg: dataset/stamp_analytics_img.bin
  binimu: dataset/stamp_analytics_imu.bin
  img: 1007, imu: 20040

rate (Hz)
  img: 25, imu: 500
sample period (s)
  img: 0.04, imu: 0.002

diff count
  imgs: 1007, imus: 20040
  imgs_t_diff: 1006, imus_t_diff: 20039

diff where (factor=0.1)
  imgs where diff > 0.04*1.1 (0)
  imgs where diff < 0.04*0.9 (0)
  imus where diff > 0.002*1.1 (0)
  imus where diff < 0.002*0.9 (0)

image timestamp duplicates: 0

save figure to:
  dataset/stamp_analytics.png
stamp analytics done

分析结果图会保存进数据集目录，参考如下：

另外，脚本具体选项可执行 -h 了解：

$ python tools/analytics/stamp_analytics.py -h

小技巧

录制数据集时建议 record.cc 里注释显示图像 cv::imshow()， dataset.cc 里注释存储图像 cv::imwrite() 。因为此些操作都比较耗时，可能会导致丢弃图像。换句话说就是消费赶不上生产，所以丢弃了部分图像。 record.cc 里用的 GetStreamDatas() 仅缓存最新的 4 张图像。