Search

The processes of motion target detection and tracking in most of traditional activity recognition methods are usually complicated and the application of these methods is limited. In this paper, we propose a fast violent activity recognition method based on motion vectors. First, we extract the motion vectors from compressed video data directly. Then, we analyze the features of the motion vectors in each frame and between frames, and get Region Motion Vectors descriptor (RMV). Finally, we use the Support Vector Machine (SVM) which takes the radial basis as the kernel function to classify the RMV and determine whether the violent activity exists in the video or not. Experimental results on several datasets have shown that the proposed method can detect 96.1% of the violent activities in videos (false probability is about 5.1%), and the calculation speed is very fast, which means the new method can be used in embedded systems.

Detecting fall down behavior is a meaningful work in the area of public video surveillance and smart home care, as this behavior is often caused by accident but usually trigger serious result. However, the uncertain individual behavior, the difference between different cameras, and the complexity of real application scene make the work absolutely hard. In this paper, a robust fall down behavior recognition algorithm is proposed based on the spatial and temporal analysis of the Key Area of Human Body (KAHB). Firstly, a modified ViBe method is applied to extract motion area. Then a pre-trained human body classifier combined with histogram tracking is used to locate the KAHB and extract its normalized spatial and temporal features. Finally, a SVM classifier is employed to find the fall down behavior.

Single Shot Multibox Detector (SSD) provides a powerful framework for detecting objects using a single deep neural network. The detection framework is one of the top object detection algorithms in both accuracy and speed which processes a large set of object locations sampled across an image. However, this framework does not behave well for the task of pedestrian detection since the images in popular pedestrian datasets have multiple objects occlusion problem and contain lots of small objects. In this paper, we incorporate deconvolution and downsampling unit into the SSD framework allowing detection network to recycle feature maps learned from images. The enhanced performance was obtained by changing the structure of classifier network, e.g., by replacing VGGNet with DenseNet. The contribution of this paper is a one-stage approach to compose a single deep neural network for pedestrian detection task in real-time. This approach addresses the typical difficulty of detecting different scale pedestrian at only one layer by providing a novel channel fusion. To solve small objects problem, base network has been replaced with more powerful one. This approach outperforms competing one-single methods on standard Caltech pedestrian dataset benchmark. It is also faster than all the other methods.

Finger-vein recognition is an increasingly promising biometric identification technology in terms of its high identification accuracy and prominent security performance. The main challenge faced by finger-vein recognition is the low recognition performance caused by segmentation error and local difference. To tackle this challenge, a finger-vein recognition method with modified binary tree (MBT) model is proposed in this paper. MBT model is used to describe the relationship and spatial structure of vein branches quantitatively. Based on the MBT model, four stages including rough selection, model correction, segment matching, and comprehensive judgment are presented to achieve a robust matching for finger-vein. Experiments demonstrate that the proposed method can boost the performance of finger-vein recognition that is degraded by segmentation error and local difference. While maintaining low complexity, the proposed method achieves 0.12 % equal error rate in the introduced dataset with 8,100 finger-vein images from 150 participants, which outperforms the state-of-the-art methods.

Face detection is the first step of vision-based driver fatigue detection method. Traditional face detection methods have problems of high false-detection rates and long detection times. A space-time restrained Adaboost method is presented in this paper that resolves these problems. Firstly, the possible position of a driver’s face in a video frame is measured relative to the previous frame. Secondly, a space-time restriction strategy is designed to restrain the detection window and scale of the Adaboost method to reduce time consumption and false-detection of face detection. Finally, a face knowledge restriction strategy is designed to confirm that the faces detected by this Adaboost method. Experiments compare the methods and confirm that a driver’s face can be detected rapidly and precisely.

A joint template matching algorithm is proposed in this paper to reduce the high rate of miss-detection and false-alarm caused by the traditional template matching algorithm during the process of multi-object detection. The proposed algorithm can reduce the influence on each object by matching all objects together according to the correlation information among different objects. Moreover, the rate of miss-detection and false-alarm in the process of single-template matching is also reduced based on the algorithm. In this paper, firstly, joint template is created from the information of relative positions among different objects. Then, matching criterion according to normalized cross correlation is generated for multi-object matching. Finally, the proposed algorithm is applied to the detection of watermarks in bill. The experiments show that the proposed algorithm has lower miss-detection and false-alarm rate comparing to the traditional NCC algorithm during the process of multi-object detection.