Guoci Cai
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Guoci Cai has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- Yiliang Wu
School of Ocean Information Engineering, Jimei University, Xiamen 361021, China
3 shared publications - Haiqi Yu
College of Ocean Information Engineering, Jimei University, China
2 shared publications - Jie Pan
College of Ocean Information Engineering, Jimei University, Xiamen, Fujian, China
2 shared publications - Xujie Zheng
College of Ocean Information Engineering, Jimei University, China
2 shared publications - Changshe Zhang
Xidian University, Xi'an, China
1 shared publication Chaopeng Li
College of Ocean Information Engineering, Jimei University, Xiamen, China
1 shared publication
Latest Publications
- Open Access
- 24 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 89124-89134
End-to-End 3D Human Pose Estimation Network With Multi-Layer Feature Fusion
The 3D human pose estimation is a technique used to determine the position of the human body in a three-dimensional space. This involves identifying body rotations, joint angles, and other pose-related information from image or video data. In this paper, we propose an end-to-end 3D human pose estimation network that is based on multi-level feature fusion.The network is composed of two main components. The first component utilizes the deepest features extracted by the backbone network. These features undergo initial data encoding and are then processed by the Semantic Information Extraction Module, which primarily consists of a multi-head self-attention mechanism. This module extracts deeper features, resulting in primary human body feature data. The second component focuses on the shallowest features and inputs them into the Global Information Processing Module, which performs global feature extraction.The features extracted from both components, along with the Bbox info (bounding box information), are collectively fed into the Iterative Regression Module. This module generates human pose data, which is then utilized to reconstruct and generate the human body using a human pose model. To evaluate the performance of our method, we train and test it on well-known benchmark datasets such as 3DPW, AGORA and MPII. Our method demonstrates exceptional performance, as it achieves a reduction of approximately 5.3% on the PA-MPJPE metric and approximately 5.1% on the MPJPE metric compared to the best model we referenced.
- 24 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 89124-89134
Open AccessEnd-to-End 3D Human Pose Estimation Network With Multi-Layer Feature Fusion
The 3D human pose estimation is a technique used to determine the position of the human body in a three-dimensional space. This involves identifying body rotations, joint angles, and other pose-related information from image or video data. In this paper, we propose an end-to-end 3D human pose estimation network that is based on multi-level feature fusion.The network is composed of two main components. The first component utilizes the deepest features extracted by the backbone network. These features undergo initial data encoding and are then processed by the Semantic Information Extraction Module, which primarily consists of a multi-head self-attention mechanism. This module extracts deeper features, resulting in primary human body feature data. The second component focuses on the shallowest features and inputs them into the Global Information Processing Module, which performs global feature extraction.The features extracted from both components, along with the Bbox info (bounding box information), are collectively fed into the Iterative Regression Module. This module generates human pose data, which is then utilized to reconstruct and generate the human body using a human pose model. To evaluate the performance of our method, we train and test it on well-known benchmark datasets such as 3DPW, AGORA and MPII. Our method demonstrates exceptional performance, as it achieves a reduction of approximately 5.3% on the PA-MPJPE metric and approximately 5.1% on the MPJPE metric compared to the best model we referenced.
- Open Access
- 20 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 419-423
Multi-view Video Splicing System
In traditional video surveillance systems, supervisors need to watch multiple surveillance images at the same time. Surveillance video resources are scattered and huge, making it difficult to browse them all and extract useful information from them in a timely manner. The video splicing system can be used to merge multiple images and splice and fuse images from different perspectives to form a whole image, which is convenient for personnel supervision. This article introduces a multi-view video splicing system: it consists of several video cameras. Through the GStreamer framework, video streaming is performed in real time and the pictures are transmitted into the video fusion system. In the system, SURF feature points are extracted from several pictures to simulate After the radial transformation, real-time splicing is performed on them. After comparing the APAP algorithm, Multi Blend algorithm, Stitching algorithm, and the best fusion line algorithm, the best fusion line algorithm is selected for video fusion. At the same time, considering that the system requires a large amount of data operations, if the CPU is simply used to process too many videos, the effect of the algorithm and the delay of the video are often unsatisfactory. The system uses CUDA to combine the SURF algorithm and image simulation. The radial transformation is performed on the GPU, which greatly improves the fusion effect and speed of the video. Through experiments, the system can complete the splicing task better. At the same time, compared with not using CUDA for acceleration, the frame rate is increased from 3 frames per second to 25 frames, and the display can basically achieve real-time effects.
- 20 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 419-423
Open AccessMulti-view Video Splicing System
In traditional video surveillance systems, supervisors need to watch multiple surveillance images at the same time. Surveillance video resources are scattered and huge, making it difficult to browse them all and extract useful information from them in a timely manner. The video splicing system can be used to merge multiple images and splice and fuse images from different perspectives to form a whole image, which is convenient for personnel supervision. This article introduces a multi-view video splicing system: it consists of several video cameras. Through the GStreamer framework, video streaming is performed in real time and the pictures are transmitted into the video fusion system. In the system, SURF feature points are extracted from several pictures to simulate After the radial transformation, real-time splicing is performed on them. After comparing the APAP algorithm, Multi Blend algorithm, Stitching algorithm, and the best fusion line algorithm, the best fusion line algorithm is selected for video fusion. At the same time, considering that the system requires a large amount of data operations, if the CPU is simply used to process too many videos, the effect of the algorithm and the delay of the video are often unsatisfactory. The system uses CUDA to combine the SURF algorithm and image simulation. The radial transformation is performed on the GPU, which greatly improves the fusion effect and speed of the video. Through experiments, the system can complete the splicing task better. At the same time, compared with not using CUDA for acceleration, the frame rate is increased from 3 frames per second to 25 frames, and the display can basically achieve real-time effects.
- Open Access
- 20 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 623-628
Depth Camera and UWB Positioning Data Fusion Algorithm for Indoor Picking Drones
Indoor growing is a trend in agriculture in the future. Using drones to replace manual monitoring and picking of fruits can save costs and free up manpower. Achieving precise picking by drones requires strong indoor positioning capabilities, but in the complex environment of indoor orchards, using a single sensor may produce data fluctuations. This project proposes a method to fuse depth camera data and UWB data to solve the problem of insufficient positioning performance of a single sensor: independently obtain the positioning data of the depth camera and UWB, use the Kalman filter to process the data noise, and calculate the data fluctuations degree, calculate the sensor data confidence, compare and obtain the sensor data with high confidence as the preliminary data for fusion, and finally perform Kalman filter processing on the preliminary fusion data again to obtain the final data. This fusion method achieves high-precision indoor positioning of UAVs. Through simulation experiments, compared with using depth cameras or UWB data independently, the data fusion solution improves the maximum error performance by approximately 18.05% and the minimum mean square error performance. Increased by approximately 4.76%.
- 20 October 2023
- Published by Association for Computing Machinery (ACM)
- p. 623-628
Open AccessDepth Camera and UWB Positioning Data Fusion Algorithm for Indoor Picking Drones
Indoor growing is a trend in agriculture in the future. Using drones to replace manual monitoring and picking of fruits can save costs and free up manpower. Achieving precise picking by drones requires strong indoor positioning capabilities, but in the complex environment of indoor orchards, using a single sensor may produce data fluctuations. This project proposes a method to fuse depth camera data and UWB data to solve the problem of insufficient positioning performance of a single sensor: independently obtain the positioning data of the depth camera and UWB, use the Kalman filter to process the data noise, and calculate the data fluctuations degree, calculate the sensor data confidence, compare and obtain the sensor data with high confidence as the preliminary data for fusion, and finally perform Kalman filter processing on the preliminary fusion data again to obtain the final data. This fusion method achieves high-precision indoor positioning of UAVs. Through simulation experiments, compared with using depth cameras or UWB data independently, the data fusion solution improves the maximum error performance by approximately 18.05% and the minimum mean square error performance. Increased by approximately 4.76%.