Bridging Safety Gaps in Graphics
summary
Camera app computes and tracks DPU CRC regions and compares them to CRCs reported by the display processing unit to detect frozen or corrupted outputs; diagonal corner-pixel CRCs and injected noise identify cropped or delayed frames. Scaling nondeterminism is solved by writing identical values into a 2x2 corner averaging block so GPU linear filtering yields deterministic pixels. Uses direct camera-driver access, per-frame CRC lists, and probabilities up to 4,096 IDs with false-match chances ≈1/1,024, 1/1,000,000, 1/1,000,000,000; enables a safety bridge over non-ASIL GPU/Wayland with reboot, fallback display or safe-state recovery.
AI-written summary · about
chapters (25)
- 0:00Introduction
- 0:20Speaker background
- 0:40ASIL and safety categories
- 5:34ASIL risk reduction
- 5:54Graphics pipeline workflow
- 7:19Safety gaps in pipeline
- 8:02Building safety bridges
- 9:06Telltale icon safety with CRC
- 11:38Rear view camera challenges
- 13:24Camera error and delay detection
- 16:59Deterministic scaled camera images
- 18:08CRC timing issues
- 22:47Solutions for camera risks
- 23:29Summary
- 23:55Q&A
- 24:57CRC in GStreamer pipelines
- 25:19Buffer recycling risks
- 26:24Buffer content integrity
- 27:08DPU CRC timing
- 28:11Future of safety components
- 29:40Wayland certification challenges
- 30:44ASIL failure rate definition
- 31:50CRC app-driver communication
- 32:11OS safety certification
- 33:24Closing remarks
related talks
30mQuick booting a software defined rear view camera
42mSoftware ISP FOSS Support for MIPI Cameras
34mExtending Libcamera to Provide a Fully Open Camera SoftISP Stack
39mHella-Parallel Compute with RenderScript
42mDisplay Sharing in Heterogeneous Systems Using Linux as High...
25mLibcamera: Making Complex Cameras Easy
40mDemystifying the Embedded Linux Graphics Stack: An Easy Introduction for Beginners
41mBeyond Complex Cameras: Complex Video Graphs Using PipeWire