VNPT used NVIDIA Metropolis to process and analyze multiple video feeds from strategically placed cameras around some of the most traffic-heavy intersections in major cities. Using containers included with NVIDIA AI Enterprise, the AI models were trained with NVIDIA pretrained models, frameworks, and TensorRT and integrated with DeepStream to detect, recognize, and classify objects such as cars, motorcycles, bicycles, and pedestrians. The video analytics applications are deployed at the edge with NVIDIA Jetson™ or on premises with T4 and A30 GPUs.
The edge cameras, combined with Jetson and pretrained models, were programmed to understand and interpret traffic signals, track the movements of identified objects, and predict potential incidents based on observed patterns. Multiple concurrent video streams are also processed on the backend servers, saving compute and operational costs.
Since it was their first time using the DeepStream SDK, VNPT initially had some challenges and experienced performance issues. With NVIDIA AI Enterprise licenses, they were able to engage NVIDIA Enterprise Support for help. VNPT solved their performance issues, set up an effective AI training system, and optimized the performance of their AI services with the help of NVIDIA AI experts.
“NVIDIA AI Enterprise, including NVIDIA DeepStream, has been instrumental in achieving our goals. With it, we have demonstrated how AI can be effectively harnessed to address real-world issues and improve the quality of life in our cities,” said Nguyen Tien Cuong, CEO of VNPT AI.
“VNPT received excellent support from NVIDIA experts to set up and optimize the AI models, resolve issues, and enhance infrastructure performance in an optimal manner,” explained Cao Thanh Ha, CTO of VNPT AI.