Choi Jun-ho, head of the AI Research Center, is a living witness to Intellibix. He explains that when he joined Intellibix 20 years ago, he vaguely thought, “If I proceed to do this type of inauspicious work for a very long time, won’t I develop into competitive?”
That vagueness has now develop into a reality. Intellibix’s 20 years of experience, rare within the AI industry, is now considered its biggest competitive edge.
Intellibix carried out over 200 public projects last 12 months alone. Since 2008, Intellibix has carried out public projects and introduced intelligent CCTVs to greater than 124 local governments, which is half of the entire within the country, and 21 out of 25 districts in Seoul.
Founded in 2000, Intellibix initially developed image synthesis and 3D restoration technologies. “On the time, it was a small company developing computer vision technology,” Choi recalled.
Since then, now we have introduced computer vision algorithms corresponding to motion-based evaluation systems, automatic tracking solutions, and license plate recognition, and we were capable of internalize the technology by introducing deep learning technology in 2017. The experience and data that Intellibix has accrued to date are areas which might be difficult for competitors to meet up with.
“The pc vision field is considered easy to enter because there are open source models and datasets, but in point of fact, it isn’t easy to even secure high-quality data,” he said.
He also said, “It isn’t easy to develop a vision AI service just by incorporating deep learning technology,” and “The differentiating factor is that now we have various know-hows, corresponding to incorporating augmented reality technology and graphics together with camera control technology to scale back false detections.”
“Intellibix was capable of grow steadily since it had experience introducing automatic tracking and video surveillance technology and had accrued data sets over time,” he emphasized.
Specifically, through long-term research on camera sensors, business products have been produced. Just as people can recognize objects more clearly after they deal with a selected object intimately, the technology for using fixed cameras and PTZ cameras together is Intellibix’s patented technology. PTZ cameras are cameras that may pan, tilt, and zoom.
“We launched a business product in 2018, and the appliance fields will expand significantly when deep learning is applied to this camera sensor technology,” he explained.
It was revealed that they’ve now reached the stage of developing ‘real-time judgment intelligence’. This isn’t only a core technology that may maximize the utilization of vision AI in various industries, but additionally a technology that can advance to the stage of realizing artificial general intelligence (AGI).
To be able to reproduce human vision and pondering processes, technology is required to acknowledge the form and movement of objects, in addition to the background, and to think in an integrated manner. Nonetheless, the present technology is barely at the extent of reproducing a few of the brain’s functions.
Especially in industrial settings, models can recognize and report risk aspects, but ultimately, people must determine whether to take motion. If real-time judgment intelligence is realized, it is anticipated that the inefficiencies and losses that arise during this process shall be minimized.
Currently, we’re developing AI solutions that may recognize situations, make judgments, and take appropriate actions like humans in defense, safety control, and industrial sites.
“Intellibix has developed around 60 to 70 vision models to date,” he said, adding, “We’re proud to say that our technological prowess to develop models and optimize them for service is world-class.” He added that they are going to use a mixture of 36 vision models and optimized camera sensors depending on the service purpose.
Moreover, that is in step with ‘spatial intelligence’, which is taken into account a key to AGI development. Spatial intelligence is the concept that to be able to overcome the constraints of huge language models (LLMs) that now we have only learned through words, we’d like to coach AI to see and understand the true world. It is a field that Google, Meta, xAI, and even Stanford University Professor Fei-Fei Li, the founding father of ImageNet, are currently working on.
CEO Choi Jun-ho said, “Intellibix has the real-time object recognition vision AI technology mandatory for the event stage of AGI,” and “Ultimately, we’ll develop a vision language model (VLM) that may recognize and judge like a human.”
He also emphasized, “Because the associated fee of developing and operating VLM is currently too high, real-time analytical intelligence using multiple small vision models is mandatory.” He added, “The power to activate only specific functions of AGI depending on the situation and purpose is feasible due to Intellivix’s unique data pipeline.”
Reporter Park Soo-bin sbin08@aitimes.com
