On-Sensor AI is changing how AI processing can be integrated into machine vision systems by performing inference directly on the image sensor. This can reduce data transfer, latency, and reliance on external computing resources, making it particularly relevant for applications requiring real-time processing and efficient system architectures.
The Triton Smart from LUCID Vision Labs combines a 12.3 MP Sony IMX501 CMOS intelligent vision sensor with on-sensor AI processing. The sensor incorporates an AI inference engine and 8 MB of on-sensor SRAM, allowing supported AI models to run directly on the sensor.
The camera supports AI applications including image classification and object detection. At 12.3 MP, LUCID specifies up to 7.4 FPS for object detection and 8.1 FPS for classification. When using a 4 × 4 image output, performance can reach up to 25 FPS for object detection and 30 FPS for classification.
The Triton Smart can provide the regular image, AI input tensor, and inference output tensor as camera data, while the image output can be reduced to as little as 4 × 4 pixels when appropriate for the application.
This architecture can help reduce the amount of data transferred to a host system and support local AI processing without requiring cloud connectivity.
Learn more about the Triton Smart
Contact Prolog Optics for more information about LUCID’s machine vision solutions.