Applying Rotation Gradient and Particle Filter Techniques to Real-Time Human Detection and Tracking
| Autor: | Po-Wei Cheng, 鄭柏偉 |
|---|---|
| Rok vydání: | 2012 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 100 With the advent of new technology and the innovation, human detection and tracking have become popular research topics. The scope of applications covers the security and surveillance, intelligent transportation systems, and home care systems. However, due to the complexity and changing background, people’s scale size, and occlusion problems, there are still limited practical applications. In order to improve the effect of detection and tracking, this study proposes a histogram oriented gradient method combined with particle filter to achieve real-time tracking of the human body. Detecting methods can be generally divided into three steps, namely, background construction, foreground subtracting and background updating. In reality, the design of those three stages is more complex in the dynamic environment. Therefore, we use histogram oriented gradient and support vector machine for training, building human descriptor in detection, and finding possible human in the films. Our tracking method uses particle filtering technique. We approach from particle sampling and then select color distribution as target feature. We find weights by computing Bhattacharyya coefficient between the target and candidate particles, and use the weighted average to estimate the final target location. We improve particle filter method by adding edge feature to overcome the shortcomings of using only one single color feature and achieve better tracking accuracy. The tracking error is compared by RMSE (root mean squared error). If only the color feature is considered, the tracking error is about 74.18. With the help of edge feature the error is reduced to approximately 61.84. Experimental results verify that the proposed system has higher tracking accuracy and is more robust. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|