The personal data
ResumeResumeShijie Sun, PhD. 2010-2019 B.S., PhD, Chang'an University School of Information, during which he visited the University of Western Australia, Australia. He is mainly engaged in research in the field of machine vision and artificial intelligence. He has published many academic papers in top international journals IEEE TPAMI, IEEE TITS and top international conference ECCV, and served as reviewer of CVPR, ECCV, NeurlNIPS, IEEE TITS, IEEE TIM and other journals and conferences. He has applied for 7 invention patents, released 10 open source projects, and the total number of his Stars has exceeded 500 at present. Import NewsIn 2023, we need to recruit 2 graduate students, if you are curious about AI and machine vision and eager to do something you are interested in, you can send me your resume (shijieSun@chd.edu.cn), our directions here include: lifelong learning, stereo vision, pose estimation and object detection tracking. if you can bring new directions, of course, you are also very welcome. We have a limited number of seats available, so hurry up and forecast. News
Others
Social positionResearchOpen CourseGraduate CoursesTraffic Video Analysis Undergraduate CoursesAlgorithm Design and Analysis Object Oriented Programming Linux System Research projectResearch on depth-level perception and understanding of visual objects in complex dynamic environmentsSupported by the National Natural Science Foundation of China (NSFC), Joint Program, Longitudinal, (2022-2025), ￥450,000This project is aimed at perceiving and understanding complex dynamic traffic environment, carrying out research on cross-view perception enhancement and knowledge migration theory, and intelligent driving environment perception technology.  Multi-vision Sensor Fusion for Train Obstacle Detection ProjectChair, Enterprise, Transversal, (2022-2023), ￥500,000This project develops a train machine vision system based on photoelectric fusion detection. The system uses multi-source fusion detection based on artificial intelligence algorithm to intelligently detect multiple obstacles that invade the train running limit in front of the train in multiple scenes, so as to enhance the active defense capability of rail vehicles, avoid collision between rail vehicles and obstacles within the braking distance or mitigate the damage caused by collision accidents, and guarantee the safe and reliable operation of rail vehicles.  Outstanding Doctoral Graduates ProgramHost, Central University, Longitudinal (2021-2023), ￥80,000This project focuses on the traffic perception task under monocular vision, and carries out three research aspects: 'design of multi-task deep network-based target detection, 3D matching and model pose estimation method' and 'design of 3D convolution-based multi-target equation of motion estimation network' around 3D model multi-view feature extraction, multi-target detection and 3D model association matching.  Real-time Spatio-Temporal Reconstruction of Large-Scale Traffic Targets Based on Multi-Camera LinkageParticipation, National Natural Science Foundation of China, ￥570,000, (2021-2024, 62072053)Based on a unified spatio-temporal reference, this project establishes the correlation between multiple cameras in a unified spatio-temporal context through automatic spatial calibration of cameras and time-synchronous calibration, and finally constructs a spatio-temporal reconstruction map describing the static attributes of traffic targets and dynamic movements in a wide range  A fast 3D traffic scene reconstruction method based on multi-target pose and motion parameter estimation with a single cameraChair, National Youth Science Foundation, ￥240,000, (2021-2023, 62006026)To rapidly and accurately reconstruct 3D traffic simulation scenes corresponding to real traffic scenes, this project focuses on the following studies based on common cameras and 3D vehicle models: constructing 3D static traffic simulation scenes corresponding to real traffic scenes, designing code-decode networks, establishing 3D vehicle model libraries, multi-view camera pose libraries, matching feature libraries and external reference feature libraries; designing multi-task-based deep learning networks, integrating The multi-task deep learning network is designed to integrate multi-target detection, 3D matching and multi-target pose estimation, estimate the 3D model and six degrees of freedom pose corresponding to the detected targets in a single image, and reconstruct the instantaneous 3D dynamic traffic simulation scene; taking video clips as the research object, the 3D convolution-based multi-target 3D motion estimation network is designed to estimate the 3D model, motion trajectory, pose motion parameters, etc. corresponding to each detected target in the video clips, and reconstruct the 3D traffic simulation scene in time clips. The 3D dynamic traffic simulation scene is reconstructed in the time clip. Combined with the static traffic scenes, a 3D traffic simulation environment consistent with the real traffic scenes is realized, and the link between reality and simulation is established. Through the research, the preliminary technical architecture of fast 3D traffic scene reconstruction based on multi-target pose and motion parameter estimation under single camera is established.  轻瓦斯图像识别设备研发项目Chair, ￥50,000, (2021.05 - 2021.12)Traditionally, the scale of the gas relay is read manually, which is effective and desirable within a certain range, such as when encountering extreme weather, inconvenient installation location or requiring 24h continuous monitoring, which greatly increases the work difficulty and intensity of the inspection staff. In addition, for the gas relay used in the field only alarm, blocking signal issued, oil level data can not be remote transmission to the background, can not judge the signal issued time and other problems, this project in the gas relay observation hole outside the installation of image sensors; and design mechanical device, the development of control software and recognition algorithm, accurate identification of the liquid level position. The system includes: mechanical device, circuit board (image hardware platform), control software, and recognition algorithm.  Mechanical instrument automatic identification and intelligent calibration equipment research and development projectChair，￥100,000，（2021-2023）This project addresses the problem of automatic recognition of mechanical meters and builds a set of algorithm and hardware framework for dial pointer recognition based on RGB camera, including: mechanical device, circuit board, control software and recognition algorithm components. RGB-D camera based bus passenger flow automatic identification system R&D projectParticipation, ￥800,000, (2014-2019)This project addresses the problem of not being able to accurately obtain the number of people in the vehicle during bus operation, and develops an automatic pedestrian boarding and alighting statistics system based on RGB-D cameras, which includes: sensing equipment (i.e. RGB-D cameras) processing equipment, pedestrian head detection algorithm, tracking algorithm, trajectory analysis algorithm, upper-end software and back-end data analysis software. Thesis2022[5] XianFeng Han, XY Huang, ShiJie Sun*,MJ Wang. “3DDACNN: 3D dense attention convolutional neural network for point cloud based object recognition. Artificial Intelligence Review. (Accepted, 中科院二区，IF=8.13) 2021 [4] Lionel Rakai, Huansheng Song, Shijie Sun*, Wentao Zhang, and Yanni Yang. “Data Association in Multiple Object Tracking: A Survey of Recent Techniques”, Expert Systems With Applications.Volume 192, 2022,116300,ISSN: 957-4174. [ [3] ShiJie Sun, Naveed Akhtar, HuanSheng Song*, Ajmal Mian, and Mubarak Shah. “Deep Affinity Network for Multiple Object Tracking.” IEEE Transactions on Pattern Analysis and Machine Intelligence. vol. 43, no. 1, pp. 104-119, 1 Jan. 2021. [ 2020[2] ShiJie Sun, Naveed Akhtar, XiangYu Song, HuanSheng Song, Ajmal Mian, and Mubarak Shah. “Simultaneous Detection and Tracking with Motion Modelling for Multiple Object Tracking.” Proceedings of the European conference on computer vision (ECCV 2020).[ 2019[1] Shijie Sun, Naveed Akhtar, Huansheng Song, Chaoyang Zhang, Jianxin Li, and Ajmal Mian. “Benchmark Data and Method for Real-Time People Counting in Cluttered Scenes Using Depth Sensors.” IEEE Transactions on Intelligent Transportation Systems 20 (10): 3599–3612. [ Technological AchievementsHonor RewardWork experience 2019.12 - now, Chang'an University
|
