To solve the problem existing in the stereo matching of low-light images, this paper proposes a stereo matching with reconstruction network based on pyramid stereo matching network(PSMNet) and reconstruction module. In view of the characteristics of the low-light image with severe and complex noise, the image reconstruction module is added into the traditional stereo matching network for automatic denoising. In this process, the image reconstruction module assists the stereo matching module in model training, so as to reduce the influence of noise on stereo matching and obtain more accurate results. The proposed method has been evaluated and achieves good performance on the Middlebury dataset which is preprocessed. In addition, a low-light binocular platform is built to get the true low-light image and test our network in night environment, results show the disparity maps are more accurate compared with previous methods.
{"title":"A Stereo Matching with Reconstruction Network for Low-light Stereo Vision","authors":"Rui Tang, Geng Zhang, Xuebin Liu","doi":"10.1145/3372806.3372821","DOIUrl":"https://doi.org/10.1145/3372806.3372821","url":null,"abstract":"To solve the problem existing in the stereo matching of low-light images, this paper proposes a stereo matching with reconstruction network based on pyramid stereo matching network(PSMNet) and reconstruction module. In view of the characteristics of the low-light image with severe and complex noise, the image reconstruction module is added into the traditional stereo matching network for automatic denoising. In this process, the image reconstruction module assists the stereo matching module in model training, so as to reduce the influence of noise on stereo matching and obtain more accurate results. The proposed method has been evaluated and achieves good performance on the Middlebury dataset which is preprocessed. In addition, a low-light binocular platform is built to get the true low-light image and test our network in night environment, results show the disparity maps are more accurate compared with previous methods.","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"49 10","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133557648","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Lu Wang, Xiaorui Liu, Xiaoqing Hu, Luyang Guan, Ming Bao
The blade-vortex interaction (BVI) is a typical helicopter noise, and has received significant attentions in the fields of structural stealth and acoustic detection. In this paper, a hybrid scheme combining aerodynamic and acoustic analysis based on the deep neural network (DNN) is proposed to achieve a better understanding of the BVI. Meanwhile, the DNN-based scale feature model (DNN-SFM) is constructed to describe the end-to-end relationship between the aero-acoustic parameters of the BVI signal and the optimal wavelet scale feature by the MZ-discrete wavelet transform. Two novel methods based on DNN-SFM are proposed for the BVI signal detection and principal component extraction, which effectively reduces the time complexity and improves the robustness in a variety of noisy environments compared to traditional algorithms. The extensive experiments on simulated and realistic data verify the effectiveness of our methods.
{"title":"Deep Neural Network-Based Scale Feature Model for BVI Detection and Principal Component Extraction","authors":"Lu Wang, Xiaorui Liu, Xiaoqing Hu, Luyang Guan, Ming Bao","doi":"10.1145/3372806.3372813","DOIUrl":"https://doi.org/10.1145/3372806.3372813","url":null,"abstract":"The blade-vortex interaction (BVI) is a typical helicopter noise, and has received significant attentions in the fields of structural stealth and acoustic detection. In this paper, a hybrid scheme combining aerodynamic and acoustic analysis based on the deep neural network (DNN) is proposed to achieve a better understanding of the BVI. Meanwhile, the DNN-based scale feature model (DNN-SFM) is constructed to describe the end-to-end relationship between the aero-acoustic parameters of the BVI signal and the optimal wavelet scale feature by the MZ-discrete wavelet transform. Two novel methods based on DNN-SFM are proposed for the BVI signal detection and principal component extraction, which effectively reduces the time complexity and improves the robustness in a variety of noisy environments compared to traditional algorithms. The extensive experiments on simulated and realistic data verify the effectiveness of our methods.","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129633489","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Dynamic movements of human skeleton have attracted more and more attention as a robust modality for action recognition. As not all temporal stages and skeleton joints are informative for action recognition, and the irrelevant information often brings noise which can degrade the detection performance, extracting discriminative temporal and spatial features becomes an important task. In this paper, we propose a novel end-to-end attention-enhanced recurrent graph convolutional network (AR-GCN) for skeleton-based action recognition. An attention-enhanced mechanism is employed in AR-GCN to pay different levels of attention to different temporal stages and spatial joints. This approach overcomes the information loss caused by only using keyframes and key joints. In particular, AR-GCN combines the graph convolutional network (GCN) with the bidirectional recurrent neural network (BRNN), which retains the irregular joints expressive power of the original GCN, while promoting its sequential modeling ability by introducing a recurrent network. Experimental results demonstrate the effectiveness of our proposed model on the widely used NTU and Kinetics datasets.
{"title":"An Attention-Enhanced Recurrent Graph Convolutional Network for Skeleton-Based Action Recognition","authors":"Xiaolu Ding, Kai Yang, Wai Chen","doi":"10.1145/3372806.3372814","DOIUrl":"https://doi.org/10.1145/3372806.3372814","url":null,"abstract":"Dynamic movements of human skeleton have attracted more and more attention as a robust modality for action recognition. As not all temporal stages and skeleton joints are informative for action recognition, and the irrelevant information often brings noise which can degrade the detection performance, extracting discriminative temporal and spatial features becomes an important task. In this paper, we propose a novel end-to-end attention-enhanced recurrent graph convolutional network (AR-GCN) for skeleton-based action recognition. An attention-enhanced mechanism is employed in AR-GCN to pay different levels of attention to different temporal stages and spatial joints. This approach overcomes the information loss caused by only using keyframes and key joints. In particular, AR-GCN combines the graph convolutional network (GCN) with the bidirectional recurrent neural network (BRNN), which retains the irregular joints expressive power of the original GCN, while promoting its sequential modeling ability by introducing a recurrent network. Experimental results demonstrate the effectiveness of our proposed model on the widely used NTU and Kinetics datasets.","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"32 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122609953","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The focus on promotion of entrepreneurial activities has been always crucial for economic development of the successful nation. Entrepreneurs are the leaders who innovate and invent ideas that give stimulus to economic growth activities. In the modern era, entrepreneurship is a key determinant of sustainable growth. Literature explains different types of entrepreneurs that are dominant in explaining the economic growth. This study investigates several inhibitors of entrepreneurship in the perpetual economy of India. Study further explores the different motivators of entrepreneurship and examines the impact of those entrepreneurial motivators on economic growth and employment. A focus group interview was conducted with entrepreneurs in 2017. Now a days advancement of technology and Artificial Intelligence (AI) has touched every sphere of our life. This paper tries to focus on the impact made by AI in entrepreneurial activities. In general, factors that enrich entrepreneurship include encouraging social entrepreneurship, improving institutional environment and supports from international organisations. For growth of the country practical implications has been identified, such as improving institutional development, creating supportive business environment with e-commerce, and promoting social entrepreneurship, security.
{"title":"Entrepreneurship and Role of AI","authors":"A. Sharma","doi":"10.1145/3372806.3374910","DOIUrl":"https://doi.org/10.1145/3372806.3374910","url":null,"abstract":"The focus on promotion of entrepreneurial activities has been always crucial for economic development of the successful nation. Entrepreneurs are the leaders who innovate and invent ideas that give stimulus to economic growth activities. In the modern era, entrepreneurship is a key determinant of sustainable growth. Literature explains different types of entrepreneurs that are dominant in explaining the economic growth. This study investigates several inhibitors of entrepreneurship in the perpetual economy of India. Study further explores the different motivators of entrepreneurship and examines the impact of those entrepreneurial motivators on economic growth and employment. A focus group interview was conducted with entrepreneurs in 2017. Now a days advancement of technology and Artificial Intelligence (AI) has touched every sphere of our life. This paper tries to focus on the impact made by AI in entrepreneurial activities. In general, factors that enrich entrepreneurship include encouraging social entrepreneurship, improving institutional environment and supports from international organisations. For growth of the country practical implications has been identified, such as improving institutional development, creating supportive business environment with e-commerce, and promoting social entrepreneurship, security.","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122626365","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
For air moving target detection with space-based radar (SBR), discrete sidelobe clutter is generally caused by strong scattering points at the sidelobe direction in the observation scene, which is difficult to discern from moving targets as a result of its strong power and special Doppler feature. To solve the above problem, the discrete clutter determination method based on filtering response loss is proposed. Firstly, the power of the potential target is calculated after clutter suppression, and then the power loss of the potential target is obtained by giving comparison to the initial power. Finally, in accordance with the criterion that power loss of discrete sidelobe clutter is high while that of the moving target is low after clutter suppression, with which the discrete sidelobe clutter can be identified based on the adaptive threshold. Simulation results with the real measured data show the feasibility and effectiveness of the proposed method.
{"title":"Discrete Sidelobe Clutter Determination Method Based on Filtering Response Loss","authors":"He Wen, Chongdi Duan, Weiwei Wang, Yu Li","doi":"10.1145/3372806.3372812","DOIUrl":"https://doi.org/10.1145/3372806.3372812","url":null,"abstract":"For air moving target detection with space-based radar (SBR), discrete sidelobe clutter is generally caused by strong scattering points at the sidelobe direction in the observation scene, which is difficult to discern from moving targets as a result of its strong power and special Doppler feature. To solve the above problem, the discrete clutter determination method based on filtering response loss is proposed. Firstly, the power of the potential target is calculated after clutter suppression, and then the power loss of the potential target is obtained by giving comparison to the initial power. Finally, in accordance with the criterion that power loss of discrete sidelobe clutter is high while that of the moving target is low after clutter suppression, with which the discrete sidelobe clutter can be identified based on the adaptive threshold. Simulation results with the real measured data show the feasibility and effectiveness of the proposed method.","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124471928","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this paper, we propose an efficient end-to-end architecture, based on Connectionist Temporal Classification (CTC), for low-resource small-footprint keyword spotting (KWS) system. For a low-resource KWS system, it is difficult for the network to thoroughly learn the features of keywords. The intuition behind our new model is that a priori information of the keyword is available. In contrast to the conventional KWS system, we modify the label set by adding the preset keyword(s) to the original label set to enhance the learning performance and optimize the final detection task of the system. Besides, CTC is applied to address the sequential alignment problem. We employ GRU as the encoding layer in our system because of the dataset small. Experiments using the WSJ0 dataset show that the proposed KWS system is significantly more accurate than the baseline system. Compared to the character-level-only KWS system, the proposed system can obviously improve the performance. Furthermore, the improved system works well in terms of low resource conditions, especially for long words.
{"title":"A Small-Footprint End-to-End KWS System in Low Resources","authors":"Gui-Xin Shi, Weiqiang Zhang, Hao Wu, Yao Liu","doi":"10.1145/3372806.3372822","DOIUrl":"https://doi.org/10.1145/3372806.3372822","url":null,"abstract":"In this paper, we propose an efficient end-to-end architecture, based on Connectionist Temporal Classification (CTC), for low-resource small-footprint keyword spotting (KWS) system. For a low-resource KWS system, it is difficult for the network to thoroughly learn the features of keywords. The intuition behind our new model is that a priori information of the keyword is available. In contrast to the conventional KWS system, we modify the label set by adding the preset keyword(s) to the original label set to enhance the learning performance and optimize the final detection task of the system. Besides, CTC is applied to address the sequential alignment problem. We employ GRU as the encoding layer in our system because of the dataset small. Experiments using the WSJ0 dataset show that the proposed KWS system is significantly more accurate than the baseline system. Compared to the character-level-only KWS system, the proposed system can obviously improve the performance. Furthermore, the improved system works well in terms of low resource conditions, especially for long words.","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125642361","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Recently, there has been an increasing interest in end-to-end speaker recognition that directly take raw speech waveform as input without any hand-crafted features such as FBANK and MFCC. SincNet is a recently developed novel convolutional neural network (CNN) architecture in which the filters in the first convolutional layer are set to band-pass filters (sinc functions). Experiments show that SincNet achieves a significant decrease in frame error rate (FER) than traditional CNNs and DNNs. In this paper we demonstrate how to improve the performance of SincNet using Multi-Task learning (MTL). In the proposed Sinc- Net architecture, besides the main task (speaker recognition), a phoneme recognition task is employed as an auxiliary task. The network uses sinc layers and convolutional layers as shared layers to improve the extensiveness of the network, and the outputs of shared layers are fed into two different sets of full-connected layers for classification. Our experiments, conducted on TIMIT corpora, show that the proposed architecture SincNet-MTL performs better than standard SincNet architecture in both classification error rates (CER) and convergence rate.
{"title":"Multi-Task Learning Based End-to-End Speaker Recognition","authors":"Yuxuan Pan, Weiqiang Zhang","doi":"10.1145/3372806.3372818","DOIUrl":"https://doi.org/10.1145/3372806.3372818","url":null,"abstract":"Recently, there has been an increasing interest in end-to-end speaker recognition that directly take raw speech waveform as input without any hand-crafted features such as FBANK and MFCC. SincNet is a recently developed novel convolutional neural network (CNN) architecture in which the filters in the first convolutional layer are set to band-pass filters (sinc functions). Experiments show that SincNet achieves a significant decrease in frame error rate (FER) than traditional CNNs and DNNs.\u0000 In this paper we demonstrate how to improve the performance of SincNet using Multi-Task learning (MTL). In the proposed Sinc- Net architecture, besides the main task (speaker recognition), a phoneme recognition task is employed as an auxiliary task. The network uses sinc layers and convolutional layers as shared layers to improve the extensiveness of the network, and the outputs of shared layers are fed into two different sets of full-connected layers for classification. Our experiments, conducted on TIMIT corpora, show that the proposed architecture SincNet-MTL performs better than standard SincNet architecture in both classification error rates (CER) and convergence rate.","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"90 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128011521","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this paper, we described the implementation of using a collection of AI services in IBM Watson to facilitate user interaction in a virtual reality space for training simulations. The project aims to increase the efficiency of training employees in an organization, by creating an immersive 3D VR environment tailored to a specific profession. Current training methods usually require an expert of the field to be hired in order to personally train these employees. The main goal of the project is to create a standard training environment which can be used and tailored by companies to train these employees without adding an additional cost.
{"title":"Implement AI Service into VR Training","authors":"J. Suttor, Julian Marin, Evan Verbus, Meng Su","doi":"10.1145/3372806.3374909","DOIUrl":"https://doi.org/10.1145/3372806.3374909","url":null,"abstract":"In this paper, we described the implementation of using a collection of AI services in IBM Watson to facilitate user interaction in a virtual reality space for training simulations. The project aims to increase the efficiency of training employees in an organization, by creating an immersive 3D VR environment tailored to a specific profession. Current training methods usually require an expert of the field to be hired in order to personally train these employees. The main goal of the project is to create a standard training environment which can be used and tailored by companies to train these employees without adding an additional cost.","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115351251","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Ming-Jen Chang, Jih-Tang Hsieh, C. Fang, Sei-Wang Chen
This study presents a vision-based human action recognition system using a deep learning technique. The system can recognize human actions successfully when the camera of a robot is moving toward the target person from various directions. Therefore, the proposed method is useful for the vision system of indoor mobile robots. The system uses three types of information to recognize human actions, namely, information from color videos, optical flow videos, and depth videos. First, Kinect 2.0 captures color videos and depth videos simultaneously using its RGB camera and depth sensor. Second, the histogram of oriented gradient features is extracted from the color videos, and a support vector machine is used to detect the human region. Based on the detected human region, the frames of the color video are cropped and the corresponding frames of the optical flow video are obtained using the Farnebäck method (https://docs.opencv=.org/3.4/d4/dee/ tutorial_optical_flow.html). The number of frames of these videos is then unified using a frame sampling technique. Subsequently, these three types of videos are input into three modified 3D convolutional neural networks (3D CNNs) separately. The modified 3D CNNs can extract the spatiotemporal features of human actions and recognize them. Finally, these recognition results are integrated to output the final recognition result of human actions. The proposed system can recognize 13 types of human actions, namely, drink (sit), drink (stand), eat (sit), eat (stand), read, sit down, stand up, use a computer, walk (horizontal), walk (straight), play with a phone/tablet, walk away from each other, and walk toward each other. The average human action recognition rate of 369 test human action videos was 96.4%, indicating that the proposed system is robust and efficient.
{"title":"A Vision-based Human Action Recognition System for Moving Cameras Through Deep Learning","authors":"Ming-Jen Chang, Jih-Tang Hsieh, C. Fang, Sei-Wang Chen","doi":"10.1145/3372806.3372815","DOIUrl":"https://doi.org/10.1145/3372806.3372815","url":null,"abstract":"This study presents a vision-based human action recognition system using a deep learning technique. The system can recognize human actions successfully when the camera of a robot is moving toward the target person from various directions. Therefore, the proposed method is useful for the vision system of indoor mobile robots. \u0000 The system uses three types of information to recognize human actions, namely, information from color videos, optical flow videos, and depth videos. First, Kinect 2.0 captures color videos and depth videos simultaneously using its RGB camera and depth sensor. Second, the histogram of oriented gradient features is extracted from the color videos, and a support vector machine is used to detect the human region. Based on the detected human region, the frames of the color video are cropped and the corresponding frames of the optical flow video are obtained using the Farnebäck method (https://docs.opencv=.org/3.4/d4/dee/ tutorial_optical_flow.html). The number of frames of these videos is then unified using a frame sampling technique. Subsequently, these three types of videos are input into three modified 3D convolutional neural networks (3D CNNs) separately. The modified 3D CNNs can extract the spatiotemporal features of human actions and recognize them. Finally, these recognition results are integrated to output the final recognition result of human actions. \u0000 The proposed system can recognize 13 types of human actions, namely, drink (sit), drink (stand), eat (sit), eat (stand), read, sit down, stand up, use a computer, walk (horizontal), walk (straight), play with a phone/tablet, walk away from each other, and walk toward each other. The average human action recognition rate of 369 test human action videos was 96.4%, indicating that the proposed system is robust and efficient.","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"92 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124648411","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Video object tracking is an important task with a broad range of applications. In this paper, we propose a novel visual tracking algorithm based on deep activation feature maps in correlation filter framework. Deep activation feature maps are generated from convolution neural network feature maps, which can discover the important part of the tracking target and overcome shape deformation and heavy occlusion. In addition, the scale variation is calculated by another correlation filter with histogram of oriented gradient (HoG) features. Moreover, we integrate the final tracking result in each frame based on the appearance model and scale model to further boost the overall tracking performance. We validate the effectiveness of our approach on a challenging benchmark, where the proposed method illustrates outstanding performance compared with the state-ofthe-art tracking algorithms
{"title":"Deep Activation Feature Maps for Visual Object Tracking","authors":"Yang Li, Zhuang Miao, Jiabao Wang","doi":"10.1145/3297067.3297088","DOIUrl":"https://doi.org/10.1145/3297067.3297088","url":null,"abstract":"Video object tracking is an important task with a broad range of applications. In this paper, we propose a novel visual tracking algorithm based on deep activation feature maps in correlation filter framework. Deep activation feature maps are generated from convolution neural network feature maps, which can discover the important part of the tracking target and overcome shape deformation and heavy occlusion. In addition, the scale variation is calculated by another correlation filter with histogram of oriented gradient (HoG) features. Moreover, we integrate the final tracking result in each frame based on the appearance model and scale model to further boost the overall tracking performance. We validate the effectiveness of our approach on a challenging benchmark, where the proposed method illustrates outstanding performance compared with the state-ofthe-art tracking algorithms","PeriodicalId":340004,"journal":{"name":"International Conference on Signal Processing and Machine Learning","volume":"58 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114886362","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：