(1) Research on methods, architecture, tools and development environment of deep learning, fast target detection and recognition based on deep learning, significance detection and image understanding, feature extraction, research and application of middle layer of deep learning, and architecture innovation of convolutional neural network in deep learning.
(2) In cooperation with the Hong Kong Productivity Council, we have carried out research on the application of deep learning in fatigue driving detection, pedestrian detection, motor vehicle detection, lane offset detection and advanced assisted driving (ADAS), and has entered the stage of industrialization. On this basis, we further carry out in-depth driving research that directly perceives driving-related information (a high-performance autonomous driving technology proposed by Princeton University at the 2015 ICCV), which is a key technology with great potential in future autonomous driving.
(3) in the manufacturing of industrial process monitoring, grinding tool monitoring, the quality of the products and parts inspection work, at present a large number of relying on artificial (human), with the disappearance of demographic dividend, based on in-depth study of industrial machine vision technology will have great opportunities and development, to gradually replace the current based on template matching, more cumbersome and inflexible machine vision technology, make industrial machine vision more intelligent and low-cost.
(4) In the aspect of intelligent transportation, the application of deep learning in virtual coil, traffic flow statistics, population statistics, illegal lane change, illegal parking and other aspects is studied. The license plate automatic recognition technology based on video developed by us has reached the best level in the industry, including license plate detection, location, character segmentation and recognition, etc., which is one of the most core technologies in intelligent transportation.
(5) Dynamic target detection, tracking and behavior analysis in intelligent monitoring. In 1997, the Defense Advanced Research Projects Agency established Carnegie Mellon University, Visual Surveillance and Monitoring (VSAM), a major Visual Surveillance project participated by MIT and other universities, mainly researches automatic video understanding technology for battlefield and common civil scene Monitoring, which is a very important technology for video Surveillance system and video management system at present. We use image processing, video analysis, and pattern recognition technology to carry out background modeling of the scene, and on this basis, target detection, extraction, tracking, recognition and behavior analysis technology and application research in the security industry.
