Electronics

AI-based autonomous imager enables face recognition on smartphones

CEA-Leti claims to be the first autonomous imager to launch smartphones and small appliances via facial recognition.

CEA-Letty Claims to be the industry’s first autonomous imager to activate smartphones and small appliances by face recognition. This imager combines automatic exposure for all lighting conditions, motion detection, feature extraction for event-based features, and artificial intelligence-based object detection.

An autonomous imager called μWAI (micro-WAY) is 1 euro coin It also features a new read and process architecture co-designed with an optimized algorithm pipeline, which is said to offer ultra-low power wakeup modes and a compact silicon implementation for cost savings.

What is an autonomous imager? Antoine Dupret, Industrial Partnership Manager at CEA-Leti, is a new class of image sensors designed with a close combination of autonomous image acquisition and image processing. “It’s actually a visual sensor because it analyzes the scene and extracts relevant information.”

According to the Institute of Technology, this is the first smart image sensor to combine automatic exposure in all lighting conditions (to allow accurate recognition in different conditions) with a dynamic range of 88 dB, and motion detection. , Feature extraction for event-based features, and AI-based object recognition that triggers reliable identification (95% accuracy).

CEA-Leti autonomous imaging device μWAI (Image: CEA-Leti)

AI-based recognition has two important advantages. “AI has two important features that are leveraged in autonomous imagers. First, AI can achieve higher recognition rates. Ultimately, it consumes by reducing the number of false alarm wakeups. We will further reduce power consumption, “says Dupre. Second, AI is somewhat versatile. This means that the same algorithm and / or hardware can be used to recognize different objects simply by changing the learned feature “weights”. Therefore, you can use the autonomous imager to recognize other objects. “

Energy saving

At the same time, due to the key features of the imager (automatic exposure for all lighting conditions, motion detection, feature extraction for event-based features, AI-based object recognition), dozens of pJ / pixel / CEA-Leti, according to existing A frame that surpasses that of off-the-shelf systems.

“The numbers are derived from power consumption, frame rate, and image sensor resolution,” Dupre said. “Given a VGA image sensor that works at 15 fps and consumes 100 mW, the energy per pixel and frame is 100e-3 / 310e3 / 15 = 21 nJ / pixel / frame without processing.”

According to researchers, a typical implementation with a low-power camera and processor requires about 10,000 times more energy than a µWAI imager.

According to Dupret, implementing COTS requires at least one image sensor and one microcontroller. “A typical low-resolution image sensor consumes tens of milliwatts. The microcontroller needs to adapt its features, such as adjusting the exposure time. Next, you need to analyze the image. Overall, the minimum power budget is in the range of hundreds of milliwatts. “

CEA-Leti promotes the imager’s 3-6 μW operation, which meets the requirements of IoT applications and runs on coin cell batteries that last for 5 years.

However, when it comes to area and volume, gains aren’t that dramatic, Dupre said. “Autonomous imagers are intended to wake up more complex systems. Therefore, more processors are needed to take advantage of images.”

Since the image is processed by the image sensor itself, the imager also provides privacy-compliant AI-based recognition. The µWAI image sensor performs data processing on the fly within the image sensor (without frame memory). Therefore, the content of the scene is not sent outside the chip, only the set of features is sent. “

Applications for µWAI image sensors include automatic switching and face recognition for mobile devices, contactless smart switching for home appliances, and sports and entertainment devices in smart homes. Smart image sensors can also be used for facial recognition, people counting, triggering alarms in smart buildings, in-vehicle status recognition, driver identification, parking status recognition, and smart unlocking systems for cars.

CEA-Leti’s team is working with STMicroelectronics to develop specific smart imager products and extend that technology to other use cases, Dupret said. Today, CEA-Leti leverages STMicroelectronics’ state-of-the-art CIS technology and BIS pixels.

“The next application depends on what our industrial partners are targeting,” said Dupre. “CEA-Leti is working with industrial partners to develop custom-made innovations.”

μWAI technology was introduced at CEA-Leti’s digital event Letty Innovation Days, June 22nd and 23rd, 2021.



https://www.electronicproducts.com/ai-based-autonomous-imager-delivers-facial-recognition-for-smartphones/ AI-based autonomous imager enables face recognition on smartphones

Back to top button