DeepX shows ultra-low-power NPU for traffic video analysis at ITS World Congress
In short
South Korean AI chip company DeepX (딥엑스) presented traffic video analysis technology built on its ultra-low-power neural processing unit (NPU) at the 2026 ITS World Congress in Gangneung, the company said on 22 October. It also signed memorandums of understanding with Korean radar firm Baida (바이다) and Vietnamese IT firm Sotatek (소타텍).
Read the full story 2 min read
South Korean AI chip company DeepX (딥엑스) presented traffic video analysis technology based on its ultra-low-power neural processing unit (NPU) at the 2026 ITS World Congress in Gangneung, Gangwon, the company said on 22 October. Yonhap described the congress as an international event where governments, companies and research institutes share technology and policy on autonomous driving, smart traffic infrastructure and road safety. [ 1 , 2 , 4 ]
According to the company, the NPU is placed in cameras and roadside equipment so that vehicles, pedestrians and unexpected incidents are analysed on site and linked to warning systems. DeepX said its NPU performs AI computation at about 1 to 5 watts and that analysing video on site reduces the amount of data and the communication burden on central servers. Asia Economy contrasted this with existing traffic cameras, which it said only record video and send it to a control centre. [ 1 , 2 , 4 ]
The company listed uses including traffic volume analysis and vehicle classification at smart intersections, pedestrian risk detection in child protection zones, and detection of accidents, congestion and wrong-way driving on highways. [ 1 , 4 ]
During the event DeepX signed a memorandum of understanding with Baida (바이다), a Korean radar company, to co-develop an AI platform that recognises the surroundings of autonomous robots and vehicles and detects objects. The two firms plan to combine Baida's radar technology with DeepX's NPU, pursue on-site demonstrations and commercialisation, and expand mass-production cooperation. DeepX also signed an MOU with Vietnamese IT company Sotatek (소타텍) on developing and mass-producing AI solutions. Asia Economy said the Vietnam effort centres on traffic, security surveillance and robots. [ 1 , 2 , 4 ]
DeepX said it is pursuing NPU use in multi-lane vehicle licence plate recognition for a smart tolling system to be applied at Korea Expressway Corporation (한국도로공사). In a Korail (코레일) railway safety AI project, it is reviewing NPU-equipped equipment to inspect train pantographs and detect trains and workers. It is also in talks with a major Korean carmaker on using commercial bus interior cameras to analyse passenger crowding and manage safety. [ 1 , 2 , 4 ]
Kang Sang-gyun (강상균), DeepX's head of domestic sales, said: "Based on our experience applying technology in transport and mobility with major Korean companies, we will expand the commercialisation of ultra-low-power AI chips in the global ITS market." Asia Economy said the company took part in the congress on the 19th and that the event runs for five days, during which it also discussed cooperation on infrastructure intelligence with overseas government officials and industry figures. [ 1 , 2 , 4 ]
Why it matters
The company says analysing video on cameras and roadside equipment can cut the data sent to central servers and the load on networks. Several applications with Korean transport operators and a carmaker are still described as being pursued, reviewed or under discussion.
Key facts
- DeepX took part in the 2026 ITS World Congress in Gangneung, Gangwon, and presented NPU-based traffic video analysis technology, the company said on 22 October. [ 1 , 2 , 4 ]
- The company said its NPU performs AI computation at about 1 to 5 watts and analyses video on site, reducing the data and communication burden on central servers. [ 1 , 2 , 4 ]
- Applications presented include traffic volume analysis and vehicle classification at smart intersections, pedestrian risk detection in child protection zones, and accident, congestion and wrong-way detection on highways. [ 1 , 4 ]
- DeepX signed an MOU with Korean radar company Baida to co-develop an AI platform that recognises the surroundings of autonomous robots and vehicles. [ 1 , 2 , 4 ]
- DeepX also signed an MOU with Vietnamese IT company Sotatek on AI solution development and mass-production cooperation. [ 1 , 2 , 4 ]
- DeepX is pursuing NPU use in multi-lane licence plate recognition for a smart tolling system to be applied at Korea Expressway Corporation (한국도로공사). [ 1 , 2 , 4 ]
- For a Korail (코레일) railway safety AI project, DeepX is reviewing NPU-equipped equipment to inspect train pantographs and detect trains and workers. [ 1 , 2 , 4 ]
Confirmed by several sources
- DeepX announced on 22 October that it presented NPU-based traffic video analysis technology at the 2026 ITS World Congress in Gangneung. [ 1 , 2 , 4 ]
- DeepX signed MOUs with Baida and Sotatek during the event. [ 1 , 2 , 4 ]
- DeepX is in talks with a major Korean carmaker on a commercial bus interior-camera solution for passenger crowding analysis and safety. [ 1 , 2 , 4 ]
Still unclear
- When DeepX presented at the congress. Asia Economy says the company took part on the 19th and that the event runs five days, while the other two outlets say only that the event is under way and the company announced on 22 October.
- Whether the Korea Expressway Corporation, Korail and carmaker projects will be adopted. The outlets describe them as being pursued, under review or under discussion, not as agreed.
- Performance of the technology. The 1 to 5 watt power figure and the reduction in data sent to servers are DeepX's own statements, and no independent test is reported.