Cheng-Bin Jin

Real-Time Human Action Recognition Using CNN Over Temporal Images for Static Video Surveillance Cameras

This paper proposes a real-time human action recognition approach to static video surveillance systems. This approach predicts human actions using temporal images and convolutional neural networks CNN. CNN is a type of deep learning model that can automatically learn features from training videos. Although the state-of-the-art methods have shown high accuracy, they consume a lot of computational resources. Another problem is that many methods assume that exact knowledge of human positions. Moreover, most of the current methods build complex handcrafted features for specific classifiers. Therefore, these kinds of methods are difficult to apply in real-world applications. In this paper, a novel CNN model based on temporal images and a hierarchical action structure is developed for real-time human action recognition. The hierarchical action structure includes three levels: action layer, motion layer, and posture layer. The top layer represents subtle actions; the bottom layer represents posture. Each layer contains one CNN, which means that this model has three CNNs working together; layers are combined to represent many different kinds of action with a large degree of freedom. The developed approach was implemented and achieved superior performance for the ICVL action dataset; the algorithm can run at around 20 frames per second.

Improvement of Accuracy for Human Action Recognition by Histogram of Changing Points and Average Speed Descriptors

Human action recognition has become an important research topic in computer vision area recently due to many applications in the real world, such as video surveillance, video retrieval, video analysis, and human-computer interaction. The goal of this paper is to evaluate descriptors which have recently been used in action recognition, namely Histogram of Oriented Gradient (HOG) and Histogram of Optical Flow (HOF). This paper also proposes new descriptors to represent the change of points within each part of a human body, caused by actions named as Histogram of Changing Points (HCP) and so-called Average Speed (AS) which measures the average speed of actions. The descriptors are combined to build a strong descriptor to represent human actions by modeling the information about appearance, local motion, and changes on each part of the body, as well as motion speed. The effectiveness of these new descriptors is evaluated in the experiments on KTH and Hollywood datasets. Category: Smart and intelligent computing

Scale-Invarinat Kernelized Correlation Filter using Convolutional Feature for Object Tracking.

Considering the recent achievements of CNN, in this study, we present a CNN-based kernelized correlation filter (KCF) online visual object tracking algorithm. Specifically, first, we incorporate the convolutional layers of CNN into the KCF to integrate features from different convolutional layers into the multiple channel. Then the KCF is used to predict the location of the object based on these features from CNN. Additionally, it is worthying noting that the linear motion model is applied when do object location to reject the fast motion of object. Subsequently, the scale adaptive method is carried out to overcome the problem of the fixed template size of traditional KCF tracker. Finally, a new tracking update model is investigated to alleviate the influence of object occlusion. The extensive evaluation of the proposed method has been conducted over OTB-100 datasets, and the results demonstrate that the proposed method achieves a highly satisfactory performance.

Occlusion-robust object tracking based on the confidence of online selected hierarchical features

In recent years, convolutional neural networks (CNNs) have been widely used for visual object tracking, especially in combination with correlation filters (CFs). However, the increasing complex CNN models introduce more useless information, which may decrease the tracking performance. This study proposes an online feature map selection method to remove noisy and irrelevant feature maps from different convolutional layers of CNN, which can reduce computation redundancy and improve tracking accuracy. Furthermore, a novel appearance model update strategy, which exploits the feedback from the peak value of response maps, is developed to avoid model corruption. Finally, an extensive evaluation of the proposed method was conducted over OTB-2013 and OTB-2015 datasets, and compared with different kinds of trackers, including deep learning-based trackers and CF-based trackers. The results demonstrate that the proposed method achieves a highly satisfactory performance.

Local Similarity Refinement of Shape-Preserved Warping for Parallax-Tolerant Image Stitching

This study proposes a local similarity refinement strategy to handle the parallax problem in image stitching. The proposed method is combined with deconvolution to acquire high-accuracy matching between corresponding source images. Shape-preserving half-projective warp was used to eliminate distortion across the non-overlapping region caused by the global projective transformation. The proposed refinement method further refines the warping result within the overlapping region, where it suppresses the parallax. The method was compared with various state-of-the-art methods: projective (global homography), AutoStitch, Zaragozas method, Zhangs method, and Changs approach. All comparisons are based on both public data sets and a proposed Inha University Computer Vision Lab (ICVL) stitching data set. The experimental results demonstrate that the proposed method is robust for handling the parallax in image stitching.