Time filter

Source Type

Guan Y.-P.,Shanghai University | Guan Y.-P.,Key Laboratory of Advanced Displays and System Application
Tien Tzu Hsueh Pao/Acta Electronica Sinica | Year: 2014

Human being daily skill can be exerted fully and bondage can be delivered efficiently in which people use ordinary equipment as an input way if pointing gesture is used for human-computer interaction(HCI). One of key problems is how to reliably recognize pointing user from HCI scene with cluttered background. A novel method has been developed based on spatio-temporal motion. According to multi-scale wavelet transform(MWT)with outstanding local characteristics both in spatial and temporal domains, it is adopted to extract foreground motion subject from cluttered scene. Some disadvantages are overcome including restrictions in environment conditions, dynamic environment variation, and a priori assumption. MWT based gradient integral graph is used to get some HOG feature vectors in pointing hand which are classified and learnt based on machine learning. Pointing user is recognized according to spatial relationship between pointing hand and its corresponding subject. Experimental results have been shown that the proposed method is efficient and viable. ©, 2014, Chinese Institute of Electronics. All right reserved.

Huang Y.,Shanghai University | Guan Y.,Shanghai University | Guan Y.,Key Laboratory of Advanced Displays and System Application
Engineering Applications of Artificial Intelligence | Year: 2015

We study the challenging problem to classify samples into a large number of classes, and propose the idea of using different Dimensionality-Reduction (DR) projections for different classes of samples. Based on this intuitive idea, the traditional Linear Discriminant Analysis (LDA) and the trace-ratio LDA are formulated to their corresponding new multi-subspace objectives. We justify that certain effects of class-adaptive feature selection are naturally achieved via our multi-subspace DR methods. Experiments on seven datasets show that, our multi-subspace trace-ratio LDA outperform its ratio-trace and single-subspace counterparts, and its advantage is more apparent when the number of classes to be classified is large. © 2015 Elsevier Ltd.

Xie S.,Shanghai University | Guan Y.,Shanghai University | Guan Y.,Key Laboratory of Advanced Displays and System Application
Multimedia Tools and Applications | Year: 2015

Automatically detecting anomaly in surveillance videos is a crucial issue for social security. Motion instability based online abnormal behaviors detection has been developed in an unsupervised way. The motion instability is composed of direction randomness and motion intensity of particles gotten by optical flow based consecutive motion feature extraction. The direction randomness is gotten based on weighted average of a circular variance of all particles. The motion intensity is determined according to average energy of all particles considering the camera perspective effect. A feature tracking based scheme has been employed to extract spatial-temporal motion features from videos to increase the processing speed. An adaptive dynamic thresholding strategy is developed to detect deviation of the track from the patterns observed both in direction randomness and motion intensity. Besides a double-threshold inference strategy is adopted to determine the range of the motion instability. A state transition model is used to reduce false alarm for confirming anomaly. The anomaly in the video is fast online detected in an ordinary hardware from a cluttered scene without any hypothesis for the scenario contents in advance. Comparative study with state-of-the-arts has indicated the superior performance of the developed approach. © 2015 Springer Science+Business Media New York

Wei K.,Shanghai University | Guan Y.-P.,Shanghai University | Guan Y.-P.,Key Laboratory of Advanced Displays and System Application
Guangdianzi Jiguang/Journal of Optoelectronics Laser | Year: 2015

A novel method is developed to automatically recognize invasion abnormality based on a 3D virtual space, aiming at some limits in abnormal invasion behavior recognition for pedestrian in a video surveillance scenario at present. A 3D plane equation is constructed according to a single pedestrian extracted and tracked by his head in a video surveillance scenario. A corresponding 3D virtual warning space in the surveillance scene is built. The problem whether a pedestrian invades warning region in a 2D scene is transformed to the one whether he intrudes the 3D virtual warning space. A sliding filter statistics strategy in ray projection of the pedestrian's head is developed to identify whether the pedestrian invades the warning protective region or not. Neither any regular shape constraint for the specified warning region or scene content previous learning is considered in the proposed approach. Some state-of-the-arts and experiments are done in some video scenes with different contents to test the performance of the proposed method in the same conditions. Experimental results show that the developed method is efficient and valid without any specific hardware support or conditional constraint for the scenario. ©, 2015, Board of Optronics Lasers. All right reserved.

Guan Y.-P.,Shanghai University | Guan Y.-P.,Key Laboratory of Advanced Displays and System Application
Proceedings - 2010 13th IEEE International Conference on Computational Science and Engineering, CSE 2010 | Year: 2010

Among gestures in non-verbal communication, pointing gesture can be taken as one of natural human computer interfaces. Vision based hand pointing is an optimal model for human-computer interaction (HCI). One of key problems among the vision based pointing gesture is how to recognize the pointing. Aiming at some limits existing in the literature, a novel method is developed to estimate pointing gestures based on some non-calibrated cameras. Multiple un-calibrated cameras are adopted to determine the pointing target based on pointing features extracted from multiple cameras and support vector machine (SVM) classifier. No explicit constraints are set on the cameras placement. Pointing user can move freely inside a wider interaction environment while pointing at some targets. The mentioned approach does not constrain the pointing surface whether is flat or not, or the target is visible by the cameras. Edge detection based on multi-scale wavelet transformation is used to extract pointing objects from a clutter background. Experiments have shown that the developed approach is efficient for pointing recognition by comparisons. © 2010 IEEE.

Discover hidden collaborations