On May 16, 2018, Rockchip released a deep learning-based target detection technology solution running on its RK3399 chip platform, which can provide a quasi-Turnkey solution for the high-end AI artificial intelligence industry, and can support both Android and Linux systems. The target detection rate reaches more than 8 frames/second.
In the field of artificial intelligence, target detection is a very popular research direction. Target detection refers to locating and classifying target objects in pictures or videos. For machines, it is difficult to directly obtain the abstract concept and positioning of objects from the RGB pixel matrix, which brings great challenges to AI artificial intelligence applications.RK3399
At present, the main research and development directions of artificial intelligence technology are: face detection, human body detection, vehicle detection, two-dimensional code detection and gesture recognition, etc., which can be widely used in monitoring, intelligent transportation, new retail, natural interaction, etc. The basis is object detection technology. The target detection technology based on deep learning has high accuracy and robustness, but the computational load is relatively large, and it cannot be practically deployed and applied in embedded devices for a long time.RK3399
In response to the AI artificial intelligence market and technical needs, Rockchip has specially optimized the MobileNet SSD network on the powerful RK3399 platform, so that the high-precision MobileNet SSD300 1.0 runs at a frame rate of more than 8 frames, and the MobileNet with slightly lower accuracy and faster speed SSD300 0.75 runs at over 11 fps. The quasi-real-time running speed brings the basic AI technology of target detection to practical use in the embedded terminal.
Rockchip's deep learning target detection technology solution based on the RK3399 chip platform can support Android or Linux systems at the same time, improve the user experience of AI products using target detection technology, greatly shorten the development cycle, and help more high-end AI smart products as soon as possible market.