srevster
Senior Member
- Joined
- Nov 10, 2020
- Messages
- 3,058
- Likes
- 5,675
A self driving vehicle has to make a decision in 10ms and react within 50ms.
an 8MP camera produces 8 million pixels per frame at 30 frames a second. Each pixel is 3 bytes for RGB. 3 bytes times 8 million pixels is 24 million bytes or 24mb per frame. There are 30 frames per second, 24mb * 30 = 960megabytes per second or 8 gigabits per second. That is 1 camera, a car will have 12. Now each stage of the pipeline has to take the incoming data and put it through a model. Each model has a 10-20ms processing time. This latency is allocated to perception not sensor fusion. Sensor fusion uses more primitive matrix multiplications using predefined transformations. So in summary not AI while driving but during the calibration stage
an 8MP camera produces 8 million pixels per frame at 30 frames a second. Each pixel is 3 bytes for RGB. 3 bytes times 8 million pixels is 24 million bytes or 24mb per frame. There are 30 frames per second, 24mb * 30 = 960megabytes per second or 8 gigabits per second. That is 1 camera, a car will have 12. Now each stage of the pipeline has to take the incoming data and put it through a model. Each model has a 10-20ms processing time. This latency is allocated to perception not sensor fusion. Sensor fusion uses more primitive matrix multiplications using predefined transformations. So in summary not AI while driving but during the calibration stage