• Machine vision is critical to future-facing products such as self-driving vehicles and delivery robots. Developing a vision-enabled machine is difficult and expensive.
• Google’s Vertex AI Vision will allow companies to build machine vision applications that can be customized to perform a specialized task.
In the summer of 1966, summer students at MIT were tasked with creating a visual system. The story was certainly true, but time distorted the event. The “summer vision project” became an urban legend often told by computer vision scientists to demonstrate how the project participants, who thought they could come up with a prototype by the end of summer, underestimated the project’s complexity.
Despite advances in hardware software and the massive collection of data and images, the problem of creating complete machine vision still eludes the industry. Artificial intelligence and machine learning have certainly brought the technology closer; why is a solution to machine vision so difficult, and how does Google’s Vertex AI Vision help bring the 1966 project to a successful completion?
Computer vision extracts information out of digital images, building algorithms that can understand the content of images and use the information for other applications. Computer vision brings together a large set of disciplines, including neuroscience, which can help computer vision by first understanding how human vision, computer science, and algorithm theory or machine learning are essential for developing computer vision algorithms. Computer vision is hard because there is a huge gap between pixels and meaning. The computer sees, in a 200 X 200 RGB image, a set of 120,000 values. The process of understanding the meaning behind the 120,000 values describes one of the core problems in computer vision.
Building a computer vision system will require a sensing device that captures as much data as possible and then transmits that data to an interpreter. This simplified vision system is how all vision systems work, both artificial and human. Where problems occur when assigning meaning to an image that requires inference and context, both of which are used to extract meaning from an image by humans. Unfortunately, AI trained with large data sets programmers to find it difficult program to infer which bits of data are relevant to identify an image correctly. Context requires prior knowledge requires massive sets of data to help a computer infer clues about an image. As image data sets grow and become more refined, these problems will be resolved, but currently, they are roadblocks in the development of accurate computer vision.
Google’s announcement regarding its Vertex AI Vision service will not solve the most vexing problems of computer vision. Instead, Vertex AI Vision takes computer vision out of the realm of the computer scientist and helps businesses manage the ingestion of data and the development and deployment of computer vision applications. AI Vision includes all the tools needed to develop computer vision applications and pre-trained models for common tasks. One of the most asked-for computer vision applications is a real-time video for security cameras that can deliver content to be reviewed after an instant has occurred. Image data can quickly grow to help manage the collected data serverless rich-media storage that provides the best of Google search combined with managed video storage. Vertex AI Vision will have the same effect as the introduction of low-code business analytics applications.