Challenge
BASF, the world’s largest chemical producer with customers in more than 190 countries, identified Computer Vision as a critical technology for addressing a significant number of its global and societal challenges. Computer Vision, the scientific endeavor that enables computers to gain high-level knowledge from digital images or videos, was a field that BASF realized would fit squarely into the developmental needs of their Coatings division. Their “Coated Object Recognition Capability” initiative was undertaken to investigate how the company could use surface color design and coating characterization techniques to enable advanced computer vision applications. Once this concept was proven viable, they searched for an external partner that not only would be highly capable in the field, but that could also provide necessary expertise in areas beyond their core competencies.
Solution
BASF’s search for an appropriate partner led to initial discussions with SRI International at an executive level. Further dialogue focused on SRI’s technological capabilities, its advanced research programs and its overall mission of helping develop and deliver innovation. After those meetings, BASF believed it had found a kindred spirit.
“We immediately saw that our competencies were quite complementary,” said Emre Kurtoglu, New Business Development Manager, Coatings Division with BASF. “With BASF’s expertise in chemicals and SRI’s expertise in taking proof of concept research to commercialization, we were more than comfortable with moving forward.”
The BASF and SRI International collaboration was established as a two-year engagement in mid-2019. To date, the joint effort is quite far along in the development of a computer vision system that includes a specialized camera and illuminator integrated with the real-time software that runs it. This system is designed to recognize thousands of BASF coatings according to their specialized color formulations, ultimately detecting and identifying a large variety objects in relation to specific tasks with unprecedented accuracy.
These color-based recognition techniques, when performed at the individual pixel level, represent significant advances in improving the overall performance of visual artificial intelligence systems. They will have direct application for smart home and mobile devices, autonomous systems, and a variety of other uses where highly accurate appearance-based decisions need to be made on manufactured objects. These advancements, resulting directly from the BASF and SRI collaboration, leverage BASF’s specialty coating formulation expertise with SRI’s color and appearance measurement technology. The practical result is a system capable of identifying a large number of specific manufactured objects in a wide variety of anticipated ambient lighting environments.
The team’s most recent achievement was the ability to properly identify approximately three-hundred BASF coatings in several typical indoor ambient light environments. Ongoing work will expand this ambient light mitigation capability to include more stressful cases, including flickering fluorescent and LED light sources. The results continue to be very promising, with 99% overall object accuracy and 95% pixel accuracy achieved so far.
According to Kurtoglu, the best part of the engagement with SRI is their ability “to be all in.” SRI has regularly illustrated its commitment to the project; showing excitement throughout the entire process, executing the work plan and ensuring all progress was properly documented. This made Kurtoglu’s job significantly easier and demonstrated SRI’s unparalleled client focus.
“I’m very happy with how SRI has engaged us,” concluded Kurtoglu. “I have worked with external partners in the past, who were very talented researchers, but their ability to properly document and report was usually inconsistent. Alternatively, SRI handled all aspects of our engagement perfectly, making it a very pleasant surprise.”