Research projects

Improvement of deep learning and feature extraction methods for enhancing performance of pattern recognition systems

[Funded by the National Foundation-NAFOSTED]

Today with the fast-growing of artificial intelligence (AI), intelligence systems have been strongly studied and developed accomplished outstanding results, such as in the areas of robotics, intelligent assistance systems, medical image-based disease diagnosis, and so on. Developed machine learning methods for increasingly feature extraction, optimal model in terms of accuracy, processing speed have become a major research trend. Some research orientations in this subdomain include developing solutions to optimize deep learning models and learning hyperparameters for high discriminated feature extraction and outperformed pattern recognition. This project concentrates on studying and proposing novel approaches in machine learning techniques and feature extraction from images. The methods are designed to provide better pattern recognition in real datasets. Resulted solutions support intelligent decisions and suggestions such as recognizing diseases using visual data in medical diagnosis, detecting abnormal objects or dangerous behaviors in surveillance systems. We expect this research to be completed and contributed to the trend machine learning solutions that facilitate implementing and operating intelligent systems on resource-limited hardware to solve difficult data problems for the classification and recognition of complex objects and events. The result is toward AI technologies into practice application with acceptable costs.

Develop AI technology based on deep learning models for feature extraction and pattern recognition to solve specific problems in computer vision such as image classification, object detection, medical image processing, action recognition, and scene understanding for surveillance systems.

Virtual tour development of famous places in Quang Binh using panoramic stitching, 3D interaction and modern web technology

[Funded by the QB People's Committee]

Development of an intelligent system for ambient understanding applied to autonomous navigation in outdoor environment.

[Funded by the National Foundation-NAFOSTED]

This project proposes and studies a novel method that improves efficiency of scene understanding, which is useful for autonomous navigation, based on the multiple sensors-based. The method is expected to provide the better scene understanding, which then allows a robot to avoid obstacles in front of its navigation. The method consists of some stages. First, a path planning is investigated, which concerns about the findings of the efficient path to facilitate the autonomous driving. The second is referred to the problem of the motion estimation and the localization prediction of running vehicle. It is addressed based on applying the fusion of cameras, LRF and GPS devices. Our research proposes a new method that uses the minimal set of parameters consisting of the geometrical constraints for estimating the 3D motion of vehicle using cameras and LRF. The cumulative errors of visual odometry are excluded using the GPS-based correction relied on the maximum likelihood estimation in particle filter. The semantic of object classification and detection are also studied. The obstacle detection, place recognition techniques are used as a solution to assisting the autonomous driver. We focus on dealing with detecting obstacles that commonly occur such as pedestrians and vehicles. The place recognition is also considered for scene understanding and mapping. We introduce improving of feature descriptors, deformable part model as well hybrid boosting SVM and neural network.

Proposal a computer vision- based surveillance system applied to secured to office building in Quang Binh province

[Funded by the QB People's Committee]

Page updated

Google Sites

Report abuse