Issue link: https://resources.mouser.com/i/1442826
9 AI ❝ ❞ I Spy With AI By Ravi Malhotra, Software Architect, NXP Digital Networking Group Which algorithm you choose for object detection depends on whether you need to detect an object or detect the right object. H ave you wondered how the automated emergency braking or pedestrian detection system in your new car works? As opposed to facial recognition, it does not care if the person in front is your mother-in-law or some stranger, it just identifies a person in the path of the car, and instructs it to apply the brakes. More advanced systems may detect additional types of objects like dogs, cats, other obstacles, and even changing road and weather conditions. Let's take a deeper look at what it takes to build such systems and what their use-cases could be. Object-detection can be broken down into two sub-categories: General object-detection, and specific object-detection. In the former, a large data-set is used to identify multiple objects within a single image, often with a lower degree of confidence. This would be useful for general image classification like the one Google does with your uploaded photos (hint: Try searching for keyword 'beer' in your Google photo library). However, in use-cases like industrial safety, the need is to detect very specific objects. For example, is the person wearing the right safety gear, like helmet, goggles, and vest? Are they carrying the right gear for the job? The degree of confidence in the detection needs to be very high as this is a critical operation, or the decision needs to be to err on side of safety. Another use- case that requires a high degree of confidence is widget quality detection. It needs to have only a single object in its data-set, but it needs to be able to detect any minor flaws in widgets coming off an assembly line in a factory, to a 99.99% or higher probability. There are multiple algorithms available for object-detection, and which one you choose depends on the level of accuracy your particular application requires. Let's take a deeper look at some of the most popular ones.