Rockchip released the deep learning-based target detection solution, speeding up the commercial process of high-end AI
Target detection is a very popular research orientation in AI field. It involves the positioning and classification of target objects in pictures or videos. It is quite difficult for machines to acquire the abstract concept of objects and position them from RGB pixel matrix, which poses great challenges to the application of AI.
Currently, the major R&D orientations of AI technology includes: face detection, body detection, vehicle detection, QR code detection and gesture detection, which can be widely used in monitoring, intelligent transportation, new retails, natural interactions, etc., while the target detection technology is the basis of these applications. The deep learning-based target detection technology is highly accurate and robust, and yet cannot be actually deployed and applied in embedded devices for a long time due to heavy computation.
Apart from the quasi real-time running speed, the solution supports the TensorFlow Lite model exported from the Google TensorFlow Object Detection training. As of now, there have been a substantial number of application cases based on TensorFlow Object Detection, covering different types of detections, from face to object, making it one of the most convenient and prevalent target detection frames in the industry.
Rockchip deep learning target detection solution, based on RK3399, supports both Android and Linux, and improves the user experience in using AI products of target detection technology. The solution substantially shortens the R&D cycle and facilitates the earlier launch of more high-end AI products.