Yueliang Qian

Crowd Density Estimation Based on Local Binary Pattern Co-Occurrence Matrix

Crowd density estimation is important for intelligent video surveillance. Many methods based on texture features have been proposed to solve this problem. Most of the existing algorithms only estimate crowd density on the whole image while ignore crowd density in local region. In this paper, we propose a novel texture descriptor based on Local Binary Pattern (LBP) Co-occurrence Matrix (LBPCM) for crowd density estimation. LBPCM is constructed from several overlapping cells in an image block, which is going to be classified into different crowd density levels. LBPCM describes both the statistical properties and the spatial information of LBP and thus makes full use of LBP for local texture features. Additionally, we both extract LBPCM on gray and gradient images to improve the performance of crowd density estimation. Finally, the sliding window technique is used to detect the potential crowded area. The experimental results show the proposed method has better performance than other texture based crowd density estimation methods.

Fast commercial detection based on audio retrieval

Automatic detection of commercials in digital multimedia material is a challenging task with many applications. This paper presents a novel approach to fast commercial detection based on audio retrieval. It is based on the idea of segmenting energy envelope of audio into units, using only audio signal for matching on a commercial database. Fast searching and matching can be performed with high accuracy, by searching and by novel similarity function based on units. Experimental results show that 96.8% recall rate and 98.7% precision rate can be achieved under 0.125 real-time.

Segment and Label Indoor Scene Based on RGB-D for the Visually Impaired

The growing study in RGB-D sensor and 3D point cloud have made new progress in obstacle avoidance for the visually impaired. However, it remains a challenging problem due to the difficulty in design a robust and real-time algorithm. In this paper, we focus on scene segmentation and labeling. As man-made indoor scene contains many planar area and structure, plane segmentation and classification is important for further scene analysis. This work propose a multiscale-voxel strategy to reduce the effects of noise and improve plane segmentation. Then the segmentation result is combined with depth data and color data to apply graph-based image segmentation algorithm. After that, a cascaded decision tree is trained to classify different segments into different semantical type. The method is tested on part of the NYU Depth Dataset. Experimental results show that the proposed method combines the advantages of depth data and the geometry characteristics of the scene, and improves scene segmentation and obstacle detection.

A ring-shaped interactive device for large remote display and mobile device control

In this demonstration, a novel human-computer interaction device is proposed to realize finger touching for large display and mobile device control, without a touchscreen or a touch pad. In this method, interaction commands are input in a same way as traditional touchscreen and touchpad, which is convenient to develop applications for long-distance operation of display and mobile devices. An embedded module is designed to collect bone-conducted friction sound, acceleration and gyroscope sensor data, corresponding to the behavior and direction of interaction commands. For algorithm, modified MFCC and SVM are applied in sound processing and probability calculation.

A Fast and Robust Pedestrian Detection Framework Based on Static and Dynamic Information

With the powerful development of pedestrian detection technique based on sliding-window and machine-learning, detection-based tracking systems have become increasingly popular. Most of these systems rely on existing static pedestrian detectors only despite the obvious potential motion information for people detection. This paper proposes a novel pedestrian detection framework fusing static and dynamic features. Motion cue is firstly used to detect potential pedestrian regions. Secondly, static detector scans potential regions to get candidate pedestrian detections. Final detection results are improved by removing false detections based on their motion distribution. The proposed framework significantly raises detection speed and detection performance. Static detector of pedestrian in this paper is trained by AdaBoost with simplified HOG feature (1HOG). Additionally, we introduce a detection-window-pyramid based scanning strategy for quickly extracting 1HOG features. The experimental results on several public data sets show the effectiveness of the proposed approach.

Related HOG Features for Human Detection Using Cascaded Adaboost and SVM Classifiers

Robust and fast human detection in static image is very important for real applications. Although different feature descriptors have been proposed for human detection, for HOG descriptor, how to select and combine more distinguish block-based HOGs, and how to simultaneously make use of the correlation and the local information of these selected HOGs still lack enough research and analysis. In this paper, we present a set of Related HOG (RHOG) features, including distinctive block-based HOGs (Ele-HOGs) which are selected by Adaboost and a global HOG descriptor which is concatenated by Ele-HOGs (CSele-HOG). Ele-HOG can discriminatively describe local distribution of human object while CSele-HOG contains global information. In addition, we propose a novel human detection framework of Cascaded Adaboost and SVM classifiers (CAS) based on RHOG features, which combines the advantages of Adaboost and SVM classifiers. Experimental results on INRIA dataset demonstrate the effectiveness of the proposed method.

A novel method for people and vehicle classification based on Hough line feature

In this paper, we propose a novel and simple method for people and vehicles classification in far distance video surveillance. In this approach, moving objects are firstly segmented from background using a background subtraction technique. Secondly, edges of moving objects are extracted using canny operator. Then straight lines of edges of moving objects are extracted by Hough transform and feature based on Hough line feature (HouLR) for classification is constructed. Finally, moving objects are classified into people or vehicle by HouLR feature. We test our method on several videos in different scenes. The experimental results show that our approach is simple and fast, and has high classification accuracy, not only can distinguish single person from vehicle but also can distinguish group of people from vehicle. The proposed method needs no advance scene calibration, no object tracking and no sample training, which is easy to transplant to other scene.

iSee: obstacle detection and feedback system for the blind

This paper proposes a novel obstacle detection and feedback system iSee based on RGB-D sensor for the blind. Firstly, a fast and robust ground detection method based on multi-scale voxel plane segmentation and geometric constraints is introduced. Secondly, the regions besides the ground area are segmented by region growing. Then an area-division based obstacle detection method is proposed, which combines the analysis of context information. Finally, a multi-level voice feedback strategy is proposed. The experimental results show the proposed system can detect and feedback obstacle information fast and robustly.

A Novel Multi-Feature Descriptor for Human Detection Using Cascaded Classifiers in Static Images

Combining multiple kinds of features is useful to achieve the state of the art performance for human detection. But combining more features will result in high dimensional feature descriptors, which is time-consuming for feature extraction and detection. How to exploit different kinds of features and reduce the dimension of feature descriptor are challenging problems. A novel multi-feature descriptor (MFD) combining Optimal Histograms of Oriented Gradients (OHOG), Local Binary Patterns (LBP) and Color Self-Similarity in Neighbor (NCSS) is proposed. Firstly, a discriminative feature selection and combination strategy is introduced to obtain distinctive local HOGs and construct OHOG feature. OHOG combines local discriminative and correlated information, which improves the classification performance compared with HOG. Besides, LBP describes texture feature of human appearance. Finally, a compact and lower dimensional feature NCSS is proposed to encode the self-similarity of color histograms in limited neighbor sub-regions instead of global regions. The proposed MFD describes human appearance from gradient, texture and color features, which can complement each other and improve the robustness of human description. To further improve detection speed without decreasing accuracy, we cascade early stages of Adaboost based on selected local HOGs and SVM classifier based on MFD. The former part can reject most non-human detection windows quickly and the final SVM classifier can guarantee a high accuracy. Experimental results on public dataset show that the proposed MFD and cascaded classifiers framework can achieve promising results both in accuracy and detection speed.

Residential Load Pattern Analysis for Smart Grid Applications Based on Audio Feature EEUPC

The smart grid is an important application field of the Internet of things. This paper presents a method of user electricity consumption pattern analysis for smart grid applications based on the audio feature EEUPC. A novel similarity function based on EEUPC is adapted to support clustering analysis of residential load patterns. The EEUPC similarity exploits features of peaks and valleys on curves instead of directly comparing values and obtains better performance for clustering analysis. Moreover, the proposed approach performs load pattern clustering, extracts a typical pattern for each cluster, and gives suggestions toward better power consumption for each typical pattern. Experimental results demonstrate that the EEUPC similarity is more consistent with human judgment than the Euclidean distance and higher clustering performance can be achieved for residential electric load data.