The Current · Future Electronics

Making Image Sensor Implementation Easier for Facial and Object Recognition

·24 min·2 clips

Watch Don Gunn plug in the Cyclops board and stream live video at 5.3 frames per second from a monochrome sensor.

Can engineers build camera enabled embedded systems themselves? Todd Baker starts with the old embedded design picture: a few sensors, serial buses, and user inputs gave a microcontroller enough context to act. That is not always enough now. Newer systems need to read the surrounding environment more directly, so camera input and machine vision start showing up in design choices much earlier. Don Gunn answers from the implementation side. The conversation stays close to the work of getting a camera module useful on an eval board, where driver availability can make or break the schedule. Todd notes that picking a sensor without a ready driver can burn a lot of time. Don knows the pain. In this system, the driver code is already prepared as proof of concept support for evaluating the image sensor. The stream stays deliberately plain. Because the sensor is raw and there is no ISP in the path, engineers get Bayer data and choose what happens next. They can route video to a display, compress it, or send it over Ethernet, depending on the application. The useful part is the scar tissue. Don's first image sensor integration took a couple of months of trial, adjustment, and making the pieces cooperate. The second took about a day. That is the center of the episode: captured integration knowledge can change schedule risk more than the sensor choice itself. The Cyclops board gives the discussion something concrete to point at. By the end, Todd ties the demo back to helping customers reach production with a better fitting, lower cost solution.

As heard by us

A practical look at camera-enabled embedded proof-of-concepts and first sensor bring-up.

The episode treats camera-enabled embedded work as a hands-on proof of concept, not a theory exercise. It follows the practical grind of finding a usable driver, dealing with raw Bayer data, and choosing whether the result belongs on a display or out over Ethernet.

Read the full review in PlayNext →

Why you'd press play

Need a realistic look at camera sensor bring-up on an embedded board?

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from The Current