foreword
Recently, I am doing application development on robots. Because the renminbi has plummeted, the price of Realsense's depth camera has skyrocketed. I have no choice but to evaluate the
domestic RGBD depth camera. Currently, I have evaluated 2 models, namely InuitiveR132 and FHL-D435i.
PS : FHL-D435I will be launched on ARM later, so far there is no official release of ARM to download tfp.
Evaluation Direction:
- Noise and motion blur for color data
It involves feature cutting, ROI, and uses color data, so there are certain requirements for color data. The illumination requirement for evaluation is [0.5, 20000]Lux, and 0.5Lux is RealsenseD435 as a benchmark. Under this illumination, the performance of color data and depth data All can be received, low-light images have little noise, and there are basically few "holes" in depth data.
1) For color images
Realsense > FHL-D435i > R132
Realsense automatic exposure uses a limited maximum gain, and the maximum gain is relatively smaller than that of R132. FHL’s camera is for debugging and secondary packaging, to be evaluated, and the debugging situation will be released later.
- Depth data "holes"
2) Depth information
Realsense > R132
R132 has serious depth information holes under low illumination, which leads to the optimization of low-illumination fill light method, which has certain risks, especially for highlight segmentation (license plate, etc.)
- Depth aligned color timestamp sync, CPU consumption.
3) Depth aligned color timestamp synchronization, CPU consumption.
Under the premise of the same frame rate and color depth alignment, R132 can achieve timestamp alignment (800 * 600) within 40ms, while Realsense can only achieve 120ms (640*480).
- Resource release for hotplug and exception handling.
4) Hot swapping and exception handling to release resources
Realsense > R132 > FHL
When the process exits abnormally, FHL restarts the process due to the release, which will cause the problem that the stream cannot be pulled. What can be done currently is when a soft interrupt is received ( SIGSEGV, INT, etc.) signal, do a resource release. May need further optimization.
PS: Because the MVS line scan camera is also used (the FHL depth camera has been connected to Hikvision’s API, and the same Hikvision library is used, the purpose is for visualization), but it exits abnormally, and the API will be reset when it is initialized again. Automatic release.
For the time being, write here first, and record it in an essay. We can discuss it together.