Issue link: https://resources.mouser.com/i/1442826
AI that describe the face numerically, which allows the face to be classified (identified) and compared against other data culled from previously detected faces. Processors for Facial Recognition Applications The sort of processing power you employ for your facial recognition application is determined by the amount of performance you require (in other words, the acceptable inference latency time), and the amount of memory available. Low-bandwidth facial recognition applications like a video door lock can often be handled by low-power, cost-effective microcontroller units such as NXP's i.MX RT series. If you need a low-power solution, but can still run a Linux® operating system, the i.MX 7ULP application processor might be the right fit for you. As the complexity of your application increases, you might shift up to NXP's high- performance i.MX 8M or Layerscape application processor families, with integrated graphics processing units (GPUs), multicore computer processing unit (CPU) cores, and digital signal processing (DSP) functionality, giving the opportunity to perform heterogeneous computing and multiple machine learning algorithms in parallel. Here's Looking at You At the end of Casablanca (1942), Humphrey Bogart looks at Ingrid Bergman and says, "Here's looking at you kid" in one of Hollywood's most quoted and romantic movie lines. Like Humphrey Bogart's character, the machines in our future are going to be able to look at us and recognize us. If Humphrey Bogart was around today, he might quip, "Of all the technologies in all the world, Artificial Intelligence and Machine Learning is about to walk into ours." ■ The process of facial recognition can be broken down into two main steps: 1. Image selection: Where a human sees a face, recognition technology sees ones and zeros. First, facial recognition software uses a process known as feature pyramid networks to find potential faces within an image. The facial recognition software eliminates portions of the image that do not appear to contain certain facial features—known as landmarks—allowing the remaining portion of the image to be resized into something smaller. This allows for faster processing time since less essential portions of the image are not processed. 2. Image classification: Once a face is confirmed within the image, it is run through a trained model, or inference engine, which resides on the processing unit within the device. This inference engine computes the embeddings I.MX 8M • High quality video with full 4K UltraHD resolution and HDR • Highest levels of pro audio fidelity with more than 20 audio channels each @384KHz • Optimized for fanless operation, low thermal system cost and long battery life LEARN MOREu FACIAL AND OBJECT RECOGNITION 8