As the number and complexity of sensors in autonomous vehicles continue to rise,multimodal fusionbased object detection algorithms are increasingly being used to detect 3D environmental information,significantly advan...As the number and complexity of sensors in autonomous vehicles continue to rise,multimodal fusionbased object detection algorithms are increasingly being used to detect 3D environmental information,significantly advancing the development of perception technology in autonomous driving.To further promote the development of fusion algorithms and improve detection performance,this paper discusses the advantages and recent advancements of multimodal fusion-based object detection algorithms.Starting fromsingle-modal sensor detection,the paper provides a detailed overview of typical sensors used in autonomous driving and introduces object detection methods based on images and point clouds.For image-based detection methods,they are categorized into monocular detection and binocular detection based on different input types.For point cloud-based detection methods,they are classified into projection-based,voxel-based,point cluster-based,pillar-based,and graph structure-based approaches based on the technical pathways for processing point cloud features.Additionally,multimodal fusion algorithms are divided into Camera-LiDAR fusion,Camera-Radar fusion,Camera-LiDAR-Radar fusion,and other sensor fusion methods based on the types of sensors involved.Furthermore,the paper identifies five key future research directions in this field,aiming to provide insights for researchers engaged in multimodal fusion-based object detection algorithms and to encourage broader attention to the research and application of multimodal fusion-based object detection.展开更多
With the rapid progress of the artificial intelligence(AI)technology and mobile internet,3D hand pose estimation has become critical to various intelligent application areas,e.g.,human-computer interaction.To avoid th...With the rapid progress of the artificial intelligence(AI)technology and mobile internet,3D hand pose estimation has become critical to various intelligent application areas,e.g.,human-computer interaction.To avoid the low accuracy of single-modal estimation and the high complexity of traditional multi-modal 3D estimation,this paper proposes a novel multi-modal multi-view(MMV)3D hand pose estimation system,which introduces a registration before translation(RT)-translation before registration(TR)jointed conditional generative adversarial network(cGAN)to train a multi-modal registration network,and then employs the multi-modal feature fusion to achieve high-quality estimation,with low hardware and software costs both in data acquisition and processing.Experimental results demonstrate that the MMV system is effective and feasible in various scenarios.It is promising for the MMV system to be used in broad intelligent application areas.展开更多
3D medical image reconstruction has significantly enhanced diagnostic accuracy,yet the reliance on densely sampled projection data remains a major limitation in clinical practice.Sparse-angle X-ray imaging,though safe...3D medical image reconstruction has significantly enhanced diagnostic accuracy,yet the reliance on densely sampled projection data remains a major limitation in clinical practice.Sparse-angle X-ray imaging,though safer and faster,poses challenges for accurate volumetric reconstruction due to limited spatial information.This study proposes a 3D reconstruction neural network based on adaptive weight fusion(AdapFusionNet)to achieve high-quality 3D medical image reconstruction from sparse-angle X-ray images.To address the issue of spatial inconsistency in multi-angle image reconstruction,an innovative adaptive fusion module was designed to score initial reconstruction results during the inference stage and perform weighted fusion,thereby improving the final reconstruction quality.The reconstruction network is built on an autoencoder(AE)framework and uses orthogonal-angle X-ray images(frontal and lateral projections)as inputs.The encoder extracts 2D features,which the decoder maps into 3D space.This study utilizes a lung CT dataset to obtain complete three-dimensional volumetric data,from which digitally reconstructed radiographs(DRR)are generated at various angles to simulate X-ray images.Since real-world clinical X-ray images rarely come with perfectly corresponding 3D“ground truth,”using CT scans as the three-dimensional reference effectively supports the training and evaluation of deep networks for sparse-angle X-ray 3D reconstruction.Experiments conducted on the LIDC-IDRI dataset with simulated X-ray images(DRR images)as training data demonstrate the superior performance of AdapFusionNet compared to other fusion methods.Quantitative results show that AdapFusionNet achieves SSIM,PSNR,and MAE values of 0.332,13.404,and 0.163,respectively,outperforming other methods(SingleViewNet:0.289,12.363,0.182;AvgFusionNet:0.306,13.384,0.159).Qualitative analysis further confirms that AdapFusionNet significantly enhances the reconstruction of lung and chest contours while effectively reducing noise during the reconstruction process.The findings demonstrate that AdapFusionNet offers significant advantages in 3D reconstruction of sparse-angle X-ray images.展开更多
In order to solve difficult detection of far and hard objects due to the sparseness and insufficient semantic information of LiDAR point cloud,a 3D object detection network with multi-modal data adaptive fusion is pro...In order to solve difficult detection of far and hard objects due to the sparseness and insufficient semantic information of LiDAR point cloud,a 3D object detection network with multi-modal data adaptive fusion is proposed,which makes use of multi-neighborhood information of voxel and image information.Firstly,design an improved ResNet that maintains the structure information of far and hard objects in low-resolution feature maps,which is more suitable for detection task.Meanwhile,semantema of each image feature map is enhanced by semantic information from all subsequent feature maps.Secondly,extract multi-neighborhood context information with different receptive field sizes to make up for the defect of sparseness of point cloud which improves the ability of voxel features to represent the spatial structure and semantic information of objects.Finally,propose a multi-modal feature adaptive fusion strategy which uses learnable weights to express the contribution of different modal features to the detection task,and voxel attention further enhances the fused feature expression of effective target objects.The experimental results on the KITTI benchmark show that this method outperforms VoxelNet with remarkable margins,i.e.increasing the AP by 8.78%and 5.49%on medium and hard difficulty levels.Meanwhile,our method achieves greater detection performance compared with many mainstream multi-modal methods,i.e.outperforming the AP by 1%compared with that of MVX-Net on medium and hard difficulty levels.展开更多
In complex traffic environment scenarios,it is very important for autonomous vehicles to accurately perceive the dynamic information of other vehicles around the vehicle in advance.The accuracy of 3D object detection ...In complex traffic environment scenarios,it is very important for autonomous vehicles to accurately perceive the dynamic information of other vehicles around the vehicle in advance.The accuracy of 3D object detection will be affected by problems such as illumination changes,object occlusion,and object detection distance.To this purpose,we face these challenges by proposing a multimodal feature fusion network for 3D object detection(MFF-Net).In this research,this paper first uses the spatial transformation projection algorithm to map the image features into the feature space,so that the image features are in the same spatial dimension when fused with the point cloud features.Then,feature channel weighting is performed using an adaptive expression augmentation fusion network to enhance important network features,suppress useless features,and increase the directionality of the network to features.Finally,this paper increases the probability of false detection and missed detection in the non-maximum suppression algo-rithm by increasing the one-dimensional threshold.So far,this paper has constructed a complete 3D target detection network based on multimodal feature fusion.The experimental results show that the proposed achieves an average accuracy of 82.60%on the Karlsruhe Institute of Technology and Toyota Technological Institute(KITTI)dataset,outperforming previous state-of-the-art multimodal fusion networks.In Easy,Moderate,and hard evaluation indicators,the accuracy rate of this paper reaches 90.96%,81.46%,and 75.39%.This shows that the MFF-Net network has good performance in 3D object detection.展开更多
The spaceborne precipitation radar onboard the Tropical Rainfall Measuring Mission satellite （TRMM PR） can provide good measurement of the vertical structure of reflectivity, while ground radar （GR） has a relative...The spaceborne precipitation radar onboard the Tropical Rainfall Measuring Mission satellite （TRMM PR） can provide good measurement of the vertical structure of reflectivity, while ground radar （GR） has a relatively high horizontal resolution and greater sensitivity. Fusion of TRMM PR and GR reflectivity data may maximize the advantages from both instruments. In this paper, TRMM PR and GR reflectivity data are fused using a neural network （NN）-based approach. The main steps included are： quality control of TRMM PR and GR reflectivity data; spatiotemporal matchup; GR calibration bias correction; conversion of TRMM PR data from Ku to S band; fusion of TRMM PR and GR reflectivity data with an NN method： interpolation of reflectivity data that are below PR＇s sensitivity; blind areas compensation with a distance weighting-based merging approach; combination of three types of data： data with the NN method, data below PR＇s sensitivity and data within compensated blind areas. During the NN fusion step, the TRMM PR data are taken as targets of the training NNs, and gridded GR data after horizontal downsampling at different heights are used as the input. The trained NNs are then used to obtain 3D high-resolution reflectivity from the original GR gridded data. After 3D fusion of the TRMM PR and GR reflectivity data, a more complete and finer-scale 3D radar reflectivity dataset incorporating characteristics from both the TRMM PR and GR observations can be obtained. The fused reflectivity data are evaluated based on a convective precipitation event through comparison with the high resolution TRMM PR and GR data with an interpolation algorithm.展开更多
3D vehicle detection based on LiDAR-camera fusion is becoming an emerging research topic in autonomous driving.The algorithm based on the Camera-LiDAR object candidate fusion method(CLOCs)is currently considered to be...3D vehicle detection based on LiDAR-camera fusion is becoming an emerging research topic in autonomous driving.The algorithm based on the Camera-LiDAR object candidate fusion method(CLOCs)is currently considered to be a more effective decision-level fusion algorithm,but it does not fully utilize the extracted features of 3D and 2D.Therefore,we proposed a 3D vehicle detection algorithm based onmultimodal decision-level fusion.First,project the anchor point of the 3D detection bounding box into the 2D image,calculate the distance between 2D and 3D anchor points,and use this distance as a new fusion feature to enhance the feature redundancy of the network.Subsequently,add an attention module:squeeze-and-excitation networks,weight each feature channel to enhance the important features of the network,and suppress useless features.The experimental results show that the mean average precision of the algorithm in the KITTI dataset is 82.96%,which outperforms previous state-ofthe-art multimodal fusion-based methods,and the average accuracy in the Easy,Moderate and Hard evaluation indicators reaches 88.96%,82.60%,and 77.31%,respectively,which are higher compared to the original CLOCs model by 1.02%,2.29%,and 0.41%,respectively.Compared with the original CLOCs algorithm,our algorithm has higher accuracy and better performance in 3D vehicle detection.展开更多
The concrete aging problem has gained more attention in recent years as more bridges and tunnels in the United States lack proper maintenance. Though the Federal Highway Administration requires these public concrete s...The concrete aging problem has gained more attention in recent years as more bridges and tunnels in the United States lack proper maintenance. Though the Federal Highway Administration requires these public concrete structures to be inspected regularly, on-site manual inspection by human operators is time-consuming and labor-intensive. Conventional inspection approaches for concrete inspection, using RGB imagebased thresholding methods, are not able to determine metric information as well as accurate location information for assessed defects for conditions. To address this challenge, we propose a deep neural network(DNN) based concrete inspection system using a quadrotor flying robot(referred to as City Flyer) mounted with an RGB-D camera. The inspection system introduces several novel modules. Firstly, a visual-inertial fusion approach is introduced to perform camera and robot positioning and structure 3 D metric reconstruction. The reconstructed map is used to retrieve the location and metric information of the defects.Secondly, we introduce a DNN model, namely Ada Net, to detect concrete spalling and cracking, with the capability of maintaining robustness under various distances between the camera and concrete surface. In order to train the model, we craft a new dataset, i.e., the concrete structure spalling and cracking(CSSC)dataset, which is released publicly to the research community.Finally, we introduce a 3 D semantic mapping method using the annotated framework to reconstruct the concrete structure for visualization. We performed comparative studies and demonstrated that our Ada Net can achieve 8.41% higher detection accuracy than Res Nets and VGGs. Moreover, we conducted five field tests, of which three are manual hand-held tests and two are drone-based field tests. These results indicate that our system is capable of performing metric field inspection,and can serve as an effective tool for civil engineers.展开更多
In response to the evolving challenges posed by small unmanned aerial vehicles(UAVs),which have the potential to transport harmful payloads or cause significant damage,we present AV-FDTI,an innovative Audio-Visual Fus...In response to the evolving challenges posed by small unmanned aerial vehicles(UAVs),which have the potential to transport harmful payloads or cause significant damage,we present AV-FDTI,an innovative Audio-Visual Fusion system designed for Drone Threat Identification.AV-FDTI leverages the fusion of audio and omnidirectional camera feature inputs,providing a comprehensive solution to enhance the precision and resilience of drone classification and 3D localization.Specifically,AV-FDTI employs a CRNN network to capture vital temporal dynamics within the audio domain and utilizes a pretrained ResNet50 model for image feature extraction.Furthermore,we adopt a visual information entropy and cross-attention-based mechanism to enhance the fusion of visual and audio data.Notably,our system is trained based on automated Leica tracking annotations,offering accurate ground truth data with millimeter-level accuracy.Comprehensive comparative evaluations demonstrate the superiority of our solution over the existing systems.In our commitment to advancing this field,we will release this work as open-source code and wearable AV-FDTI design,contributing valuable resources to the research community.展开更多
基金funded by the Yangtze River Delta Science and Technology Innovation Community Joint Research Project(2023CSJGG1600)the Natural Science Foundation of Anhui Province(2208085MF173)Wuhu“ChiZhu Light”Major Science and Technology Project(2023ZD01,2023ZD03).
文摘As the number and complexity of sensors in autonomous vehicles continue to rise,multimodal fusionbased object detection algorithms are increasingly being used to detect 3D environmental information,significantly advancing the development of perception technology in autonomous driving.To further promote the development of fusion algorithms and improve detection performance,this paper discusses the advantages and recent advancements of multimodal fusion-based object detection algorithms.Starting fromsingle-modal sensor detection,the paper provides a detailed overview of typical sensors used in autonomous driving and introduces object detection methods based on images and point clouds.For image-based detection methods,they are categorized into monocular detection and binocular detection based on different input types.For point cloud-based detection methods,they are classified into projection-based,voxel-based,point cluster-based,pillar-based,and graph structure-based approaches based on the technical pathways for processing point cloud features.Additionally,multimodal fusion algorithms are divided into Camera-LiDAR fusion,Camera-Radar fusion,Camera-LiDAR-Radar fusion,and other sensor fusion methods based on the types of sensors involved.Furthermore,the paper identifies five key future research directions in this field,aiming to provide insights for researchers engaged in multimodal fusion-based object detection algorithms and to encourage broader attention to the research and application of multimodal fusion-based object detection.
文摘With the rapid progress of the artificial intelligence(AI)technology and mobile internet,3D hand pose estimation has become critical to various intelligent application areas,e.g.,human-computer interaction.To avoid the low accuracy of single-modal estimation and the high complexity of traditional multi-modal 3D estimation,this paper proposes a novel multi-modal multi-view(MMV)3D hand pose estimation system,which introduces a registration before translation(RT)-translation before registration(TR)jointed conditional generative adversarial network(cGAN)to train a multi-modal registration network,and then employs the multi-modal feature fusion to achieve high-quality estimation,with low hardware and software costs both in data acquisition and processing.Experimental results demonstrate that the MMV system is effective and feasible in various scenarios.It is promising for the MMV system to be used in broad intelligent application areas.
基金Supported by Sichuan Science and Technology Program(2023YFSY0026,2023YFH0004).
文摘3D medical image reconstruction has significantly enhanced diagnostic accuracy,yet the reliance on densely sampled projection data remains a major limitation in clinical practice.Sparse-angle X-ray imaging,though safer and faster,poses challenges for accurate volumetric reconstruction due to limited spatial information.This study proposes a 3D reconstruction neural network based on adaptive weight fusion(AdapFusionNet)to achieve high-quality 3D medical image reconstruction from sparse-angle X-ray images.To address the issue of spatial inconsistency in multi-angle image reconstruction,an innovative adaptive fusion module was designed to score initial reconstruction results during the inference stage and perform weighted fusion,thereby improving the final reconstruction quality.The reconstruction network is built on an autoencoder(AE)framework and uses orthogonal-angle X-ray images(frontal and lateral projections)as inputs.The encoder extracts 2D features,which the decoder maps into 3D space.This study utilizes a lung CT dataset to obtain complete three-dimensional volumetric data,from which digitally reconstructed radiographs(DRR)are generated at various angles to simulate X-ray images.Since real-world clinical X-ray images rarely come with perfectly corresponding 3D“ground truth,”using CT scans as the three-dimensional reference effectively supports the training and evaluation of deep networks for sparse-angle X-ray 3D reconstruction.Experiments conducted on the LIDC-IDRI dataset with simulated X-ray images(DRR images)as training data demonstrate the superior performance of AdapFusionNet compared to other fusion methods.Quantitative results show that AdapFusionNet achieves SSIM,PSNR,and MAE values of 0.332,13.404,and 0.163,respectively,outperforming other methods(SingleViewNet:0.289,12.363,0.182;AvgFusionNet:0.306,13.384,0.159).Qualitative analysis further confirms that AdapFusionNet significantly enhances the reconstruction of lung and chest contours while effectively reducing noise during the reconstruction process.The findings demonstrate that AdapFusionNet offers significant advantages in 3D reconstruction of sparse-angle X-ray images.
基金National Youth Natural Science Foundation of China(No.61806006)Innovation Program for Graduate of Jiangsu Province(No.KYLX160-781)Jiangsu University Superior Discipline Construction Project。
文摘In order to solve difficult detection of far and hard objects due to the sparseness and insufficient semantic information of LiDAR point cloud,a 3D object detection network with multi-modal data adaptive fusion is proposed,which makes use of multi-neighborhood information of voxel and image information.Firstly,design an improved ResNet that maintains the structure information of far and hard objects in low-resolution feature maps,which is more suitable for detection task.Meanwhile,semantema of each image feature map is enhanced by semantic information from all subsequent feature maps.Secondly,extract multi-neighborhood context information with different receptive field sizes to make up for the defect of sparseness of point cloud which improves the ability of voxel features to represent the spatial structure and semantic information of objects.Finally,propose a multi-modal feature adaptive fusion strategy which uses learnable weights to express the contribution of different modal features to the detection task,and voxel attention further enhances the fused feature expression of effective target objects.The experimental results on the KITTI benchmark show that this method outperforms VoxelNet with remarkable margins,i.e.increasing the AP by 8.78%and 5.49%on medium and hard difficulty levels.Meanwhile,our method achieves greater detection performance compared with many mainstream multi-modal methods,i.e.outperforming the AP by 1%compared with that of MVX-Net on medium and hard difficulty levels.
基金The authors would like to thank the financial support of Natural Science Foundation of Anhui Province(No.2208085MF173)the key research and development projects of Anhui(202104a05020003)+2 种基金the anhui development and reform commission supports R&D and innovation project([2020]479)the national natural science foundation of China(51575001)Anhui university scientific research platform innovation team building project(2016-2018).
文摘In complex traffic environment scenarios,it is very important for autonomous vehicles to accurately perceive the dynamic information of other vehicles around the vehicle in advance.The accuracy of 3D object detection will be affected by problems such as illumination changes,object occlusion,and object detection distance.To this purpose,we face these challenges by proposing a multimodal feature fusion network for 3D object detection(MFF-Net).In this research,this paper first uses the spatial transformation projection algorithm to map the image features into the feature space,so that the image features are in the same spatial dimension when fused with the point cloud features.Then,feature channel weighting is performed using an adaptive expression augmentation fusion network to enhance important network features,suppress useless features,and increase the directionality of the network to features.Finally,this paper increases the probability of false detection and missed detection in the non-maximum suppression algo-rithm by increasing the one-dimensional threshold.So far,this paper has constructed a complete 3D target detection network based on multimodal feature fusion.The experimental results show that the proposed achieves an average accuracy of 82.60%on the Karlsruhe Institute of Technology and Toyota Technological Institute(KITTI)dataset,outperforming previous state-of-the-art multimodal fusion networks.In Easy,Moderate,and hard evaluation indicators,the accuracy rate of this paper reaches 90.96%,81.46%,and 75.39%.This shows that the MFF-Net network has good performance in 3D object detection.
基金supported by funding from the Natural Science Foundation of Jiangsu Province (Grant No. BK20171457)the 2013 Special Fund for Meteorological Scientific Research in the Public Interest (Grant No. GYHY201306078)+1 种基金the National Natural Science Foundation of China (Grant No. 41301399)Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD)
文摘The spaceborne precipitation radar onboard the Tropical Rainfall Measuring Mission satellite （TRMM PR） can provide good measurement of the vertical structure of reflectivity, while ground radar （GR） has a relatively high horizontal resolution and greater sensitivity. Fusion of TRMM PR and GR reflectivity data may maximize the advantages from both instruments. In this paper, TRMM PR and GR reflectivity data are fused using a neural network （NN）-based approach. The main steps included are： quality control of TRMM PR and GR reflectivity data; spatiotemporal matchup; GR calibration bias correction; conversion of TRMM PR data from Ku to S band; fusion of TRMM PR and GR reflectivity data with an NN method： interpolation of reflectivity data that are below PR＇s sensitivity; blind areas compensation with a distance weighting-based merging approach; combination of three types of data： data with the NN method, data below PR＇s sensitivity and data within compensated blind areas. During the NN fusion step, the TRMM PR data are taken as targets of the training NNs, and gridded GR data after horizontal downsampling at different heights are used as the input. The trained NNs are then used to obtain 3D high-resolution reflectivity from the original GR gridded data. After 3D fusion of the TRMM PR and GR reflectivity data, a more complete and finer-scale 3D radar reflectivity dataset incorporating characteristics from both the TRMM PR and GR observations can be obtained. The fused reflectivity data are evaluated based on a convective precipitation event through comparison with the high resolution TRMM PR and GR data with an interpolation algorithm.
基金supported by the Financial Support of the Key Research and Development Projects of Anhui (202104a05020003)the Natural Science Foundation of Anhui Province (2208085MF173)the Anhui Development and Reform Commission Supports R&D and Innovation Projects ([2020]479).
文摘3D vehicle detection based on LiDAR-camera fusion is becoming an emerging research topic in autonomous driving.The algorithm based on the Camera-LiDAR object candidate fusion method(CLOCs)is currently considered to be a more effective decision-level fusion algorithm,but it does not fully utilize the extracted features of 3D and 2D.Therefore,we proposed a 3D vehicle detection algorithm based onmultimodal decision-level fusion.First,project the anchor point of the 3D detection bounding box into the 2D image,calculate the distance between 2D and 3D anchor points,and use this distance as a new fusion feature to enhance the feature redundancy of the network.Subsequently,add an attention module:squeeze-and-excitation networks,weight each feature channel to enhance the important features of the network,and suppress useless features.The experimental results show that the mean average precision of the algorithm in the KITTI dataset is 82.96%,which outperforms previous state-ofthe-art multimodal fusion-based methods,and the average accuracy in the Easy,Moderate and Hard evaluation indicators reaches 88.96%,82.60%,and 77.31%,respectively,which are higher compared to the original CLOCs model by 1.02%,2.29%,and 0.41%,respectively.Compared with the original CLOCs algorithm,our algorithm has higher accuracy and better performance in 3D vehicle detection.
基金supported in part by the U.S.National Science Foundation(IIP-1915721)the U.S.Department of TransportationOffice of the Assistant Secretary for Research and Technology(USDOTOST-R)(69A3551747126)through INSPIRE University Transportation Center(http//inspire-utc.mst.edu)at Missouri University of Science and Technology。
文摘The concrete aging problem has gained more attention in recent years as more bridges and tunnels in the United States lack proper maintenance. Though the Federal Highway Administration requires these public concrete structures to be inspected regularly, on-site manual inspection by human operators is time-consuming and labor-intensive. Conventional inspection approaches for concrete inspection, using RGB imagebased thresholding methods, are not able to determine metric information as well as accurate location information for assessed defects for conditions. To address this challenge, we propose a deep neural network(DNN) based concrete inspection system using a quadrotor flying robot(referred to as City Flyer) mounted with an RGB-D camera. The inspection system introduces several novel modules. Firstly, a visual-inertial fusion approach is introduced to perform camera and robot positioning and structure 3 D metric reconstruction. The reconstructed map is used to retrieve the location and metric information of the defects.Secondly, we introduce a DNN model, namely Ada Net, to detect concrete spalling and cracking, with the capability of maintaining robustness under various distances between the camera and concrete surface. In order to train the model, we craft a new dataset, i.e., the concrete structure spalling and cracking(CSSC)dataset, which is released publicly to the research community.Finally, we introduce a 3 D semantic mapping method using the annotated framework to reconstruct the concrete structure for visualization. We performed comparative studies and demonstrated that our Ada Net can achieve 8.41% higher detection accuracy than Res Nets and VGGs. Moreover, we conducted five field tests, of which three are manual hand-held tests and two are drone-based field tests. These results indicate that our system is capable of performing metric field inspection,and can serve as an effective tool for civil engineers.
基金National Research Foundation,Singapore,under its Medium-Sized Center for Advanced Robotics Technology Innovation(CARTIN)under project WP5 within the Delta-NTU Corporate Lab with funding support from A*STAR under its IAF-ICP program(Grant no:I2201E0013)and Delta Electronics Inc.
文摘In response to the evolving challenges posed by small unmanned aerial vehicles(UAVs),which have the potential to transport harmful payloads or cause significant damage,we present AV-FDTI,an innovative Audio-Visual Fusion system designed for Drone Threat Identification.AV-FDTI leverages the fusion of audio and omnidirectional camera feature inputs,providing a comprehensive solution to enhance the precision and resilience of drone classification and 3D localization.Specifically,AV-FDTI employs a CRNN network to capture vital temporal dynamics within the audio domain and utilizes a pretrained ResNet50 model for image feature extraction.Furthermore,we adopt a visual information entropy and cross-attention-based mechanism to enhance the fusion of visual and audio data.Notably,our system is trained based on automated Leica tracking annotations,offering accurate ground truth data with millimeter-level accuracy.Comprehensive comparative evaluations demonstrate the superiority of our solution over the existing systems.In our commitment to advancing this field,we will release this work as open-source code and wearable AV-FDTI design,contributing valuable resources to the research community.
