Advanced Computer Vision based Human Pose Estimation
The most emerging Computer Vision-based work is Human Pose Estimation that mainly concerned with tracking, detecting, and associating key points in the human body. These key points are disturbed when dealing with video-based data. This effect limits the accuracy of the conventional Human Pose Estima
2025-06-28 16:25:01 - Adil Khan
Advanced Computer Vision based Human Pose Estimation
Project Area of Specialization Biomedical EngineeringProject SummaryThe most emerging Computer Vision-based work is Human Pose Estimation that mainly concerned with tracking, detecting, and associating key points in the human body. These key points are disturbed when dealing with video-based data. This effect limits the accuracy of the conventional Human Pose Estimation technique. To overcome this problem, the objective of this work is to develop a Convolutional Neural Networks based algorithm. In this project, the videos and images in the dataset will be converted into multiple frames. Each frame is assigned with 25 key points based on 3 factors. So, combined 75 key-points will be used for pose estimation. Dataset can be collected from two sources i.e., data collection sources such as MPII and Microsoft COCO and the other source is from camera mainly focusing on the moving people. After acquiring the dataset this data set will be trained, tested, and evaluated on CNN based algorithm. The proposed algorithm will reduce the RMSE, MAPE and MAE of human pose. The result will be compared with the existing literature.
Project ObjectivesThe project has following objectives:
- To obtain dataset by recording videos from public places.
- Data preprocessing using appropriate techniques.
- CNN based algorithm development for pose estimation.
- Detection of poses adopted by a person during video.
- Generation of Excel Sheet containing data of each Frame.
- To extract features from collected dataset.
- To evaluate the proposed algorithm by training and testing.
- To reduce the Human Pose Errors.
- Training, testing and evaluation of the dataset.
- Performance evaluation and computational complexity analysis.
Key-points detection, Pose estimation, Joints traction, action identification, and human-computer interaction all of these can be achieved by proposed CNNs based algorithm. This algorithm will extract key-points of person from each frame of recorded videos. Then excel sheets will be prepared containing these key points of human body poses. These prepared excel sheets will then utilized to prepare a dataset which will then be used to achieve all aforesaid applications.
Benefits of the ProjectWe can monitor an object or person (or numerous persons) in real-world space at an extraordinarily detailed level using posture estimation. This versatile ability opens up a world of possibilities. Pose estimation is also distinct from other standard computer vision problems in a number of key respects. Object detection is a task that locates objects in an image or video. Pose estimation also aids us in determining the precise placement of the object's key points. In addition to tracking human movement and activity, pose estimation opens up applications in a range of areas, such as: Augmented reality, Animation, Gaming, gesture and gait recognition, Motion Capture and Augmented Reality
Technical Details of Final DeliverableIn this technique, the videos and images in the dataset will be converted into multiple frames. Each frame will consist of 25 key points based on the following three factors.
- Location on the X axis.
- Location on the Y-axis.
- The value of intensity.
Therefore, each frame will have a total 75 key points. By using these key points we can efficiently detect Human poses. And baseline techniques have limitations that make it difficult to sustain high accuracy and precision (>80%) in prediction when dealing with live or recorded videos. And human pose errors such as RMSE, MAPE and MAE occur. The proposed technique will overcome all these challenges.
Final Deliverable of the Project HW/SW integrated systemCore Industry HealthOther Industries IT , Medical Core Technology Artificial Intelligence(AI)Other Technologies RoboticsSustainable Development Goals Good Health and Well-Being for People, Industry, Innovation and Infrastructure, Life on LandRequired Resources| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 80000 | |||
| HD GoPro Cameras | Equipment | 1 | 25000 | 25000 |
| GPU System (Only) | Equipment | 1 | 40000 | 40000 |
| Video Hangers | Equipment | 1 | 5000 | 5000 |
| Prototyping and Misc. | Miscellaneous | 1 | 10000 | 10000 |