Stradvision, which develops autonomous driving software ‘SVNet’, has been participating in the World Consumer Electronics Show (CES) every year since its establishment in 2014. This year, it was an opportunity to confirm the more advanced SVNet performance.
Autonomous driving consists of three core technologies: perception, judgment, and control. In human terms, it is like seeing with the eyes, judging with the brain, and moving with the feet. SVNet, the StradVision recognition software, acts as the eyes.
Currently, SVNet has been implemented in 3D and is about to be mass-produced soon. Stradvision CEO Kim Jun-hwan emphasized, “3D recognition network is a technology that receives video input and provides information about the 3D physical space in which the actual vehicle drives as a deep learning network output.”
He continued, “The existing 2D recognition method was a deep learning network that exported information in image space (2D) and converted the 2D information into information in 3D space through post-processing. This was a post-processing technology without 3D physical information. “There were many difficulties in securing accuracy performance due to limitations,” he said. “In the end, it was determined that it would be difficult to secure performance at the mass production level with this method, and the transition to developing a 3D recognition network began.”
The 3D recognition network uses LiDAR Point (distance information) as GT (Ground Truth) from network learning to minimize separate post-processing and provide 3D physical space information required for actual autonomous driving with higher accuracy.
In addition to improving accuracy at the mass production level, the 3D recognition network is also easy to develop by expanding the input channel from a single camera to multiple cameras because it is LiDAR GT-based learning. StradVision explains that this means that the 3D recognition network development experience and output of Front Vision and Surround Vision can be expanded to multi in the future, allowing for faster and more efficient development.
“StradVision focuses on providing scalable, hardware-independent solutions that combine high performance and cost-effectiveness,” said CEO Kim. “Adaptability to a variety of systems and unmatched accuracy are what sets us apart from our competitors.” “He said.
Based on these technological strengths, combined with strong customer support, StradVision has established itself as a top-level partner in the entire industry and a preferred partner for OEMs.
As SVNet is advanced into a 3D recognition network, the accuracy of recognizing the surrounding environment is close to 100%. He said, “Basically, 100% accuracy is an impossible figure, but although it will vary depending on the item, the basic aggressive accuracy figure can be said to be 98-99%.”
Stradvision is also pursuing listing on the stock market. The strategy is to reduce the variables of listing withdrawal as much as possible as the exchange’s review system for technology special listings becomes more stringent after 2023. CEO Kim Jun-hwan said, “In the past, in the case of technology special listings, the review was conducted with a focus on technology, but after 2023, the importance of evaluating business feasibility has increased.” He added, “As of October 2024, the strengthened review policy is becoming a reality, with 30% of companies that received review application results withdrawing their listings.”
In a situation where the profitability of an ongoing project is uncertain under a strengthened review policy, it was judged that it would be best to have the sales performance reach a level that would be recognized by the exchange or earnest sales performance rather than pushing the IPO too hard and then withdrawing the listing because the target public offering price was not achieved. am.
He said, “We have completed the preparation of all matters required as a listed company, such as revising the articles of incorporation, stock split (1:100), and introduction of electronic securities, referring to the case of a KOSDAQ listed company in 2024.” He added, “The listing schedule and various consents from shareholders required for this. “We plan to complete discussions on such matters in the first quarter of 2025 and apply for technology evaluation in the second quarter,” he said. He went on to say, “After obtaining the target results, we plan to request preliminary screening without delay and receive screening approval and submit a securities report in the third quarter.” He added, “We plan to apply for listing within the year at the earliest and in February 2026 at the latest.”
Jeong Jin-su, Donga.com reporter [email protected]
-
- great
- 0dog
-
- I’m sad
- 0dog
-
- I’m angry
- 0dog