Dissertation > Excellent graduate degree dissertation topics show

Research of Object Tracking Based on Monocular Vision under Indoor Environment

Author: FanXuFeng
Tutor: LiRuiFeng
School: Harbin Institute of Technology
Course: Mechanical and Electronic Engineering
Keywords: Visual Tracking Feature matching Mean shift Particle filter Similar distance
CLC: TP242
Type: Master's thesis
Year: 2008
Downloads: 55
Quote: 2
Read: Download Dissertation

Abstract


Visual tracking is a hot research field of service robots. This paper presents a similarity-based distance tracking algorithm. The algorithm visual tracking into target identification, target tracking in two parts. The similarity distances decide to perform identification or tracking algorithms, and adjust the tracking number of particles. In this paper, based on the feature matching method to identify and target. In order to effectively describe the characteristics of a variety of image feature description method based on using better robustness SIFT (Scale Invariant Feature Transform) features are described. Get the current visual image SIFT features the BBF match the characteristics of the method and the target template match. BBF (Best Bin First) matching method has a certain error rate, so this use RANSAC (Random Sample Consensus)-based method to describe the current field image and the template image on extremely fundamental matrix geometric constraints, and BBF matching results error is removed. Eliminate remaining after matching the number exceeds a certain threshold, we think that the vision in the presence of the target, and feature coordinate the completion of positioning. The algorithm is based on the color distribution information to be tracked. Around the target position of the first previously obtained to generate a certain number of particles, evenly distributed weights. Then calculate the distribution of particles with the target template color similarity distance, and then update the value of the right of individual particles, the weighted average forecast target location. When a large number of particles, mean shift introduced tracking part. Using the mean shift algorithm to determine the location of the current visual image target. After the tracking section, to calculate the similarity distance of the target area and the template, the set threshold value and the threshold value for the mean shift of the implementation of the recognition algorithm, and to make a judgment based on this distance. If the distance is greater than the recognition threshold value, then the target disappears from the current field of vision, the need to re-execute target recognition algorithm; if the distance is less than a recognition threshold, then the tracking successful, and according to the distance on the number of particles generated when the next particle filter adjustment. When the similarity distance is less than the mean shift threshold value, the better tracking effect, then the number of particles is proportional to the similarity distance; when the similarity distance between the mean shift threshold value and the identification algorithm threshold between tracking less ideals and the need to increase the number of particles to improve the tracking performance. In order to improve the computational efficiency, using fewer particles make up for the lack of the number of particles in the particle filter and mean shift algorithm. Tracking based on the introduction of perspective projection theory, a measure the target relative orientation of the lens and the distance algorithm. Azimuth and distance for the the robot crawl goal to provide information. Finally, using the USB interface, the camera on the actual object tracking, tracking a consumption of time determined by the number of particles. A similarity threshold can be set by experiment, adjusting the number of particles, and to ensure accurate tracking at the same time improve the real-time, tracking a consumption of time of approximately 5 ~~ 10ms. Experiments show that the algorithm has good real-time, accuracy and practicality.

Related Dissertations

  1. Research on Cooperative Orbit Determination in Satellite Network Based on Multi-Agent System Theory,V474
  2. The Maneuvering Target Tracking Research Based on VRPF,TN957.52
  3. Research on the Theory of Visual Object Tracking Based on Particle Filter for Autonomous Robot,TP242
  4. Tracking Cells in High Density Image Sequences Based on Mean Shift Algorithm Combined with Topological Constraint,Q25
  5. Space Infrared Target Simulation and Application of Target Trace,TP391.41
  6. Detection and Tracking of Moving Object in Complex Background,TP391.41
  7. Research on Object Detection and Tracking Method in Active Vision System,TP391.41
  8. The Research on the Target Localization and Tracking Based on WSN,TN929.5
  9. The Study of Moving Object Detection and Tracking Algorithm Based on Image Information,TP391.41
  10. The Research on Localization and Target Tracking in Wireless Sensor Network,TN929.5
  11. Architectural Optimizations for Particle Filters,TN713
  12. Mean-Shift -based KLT and target tracking study,TP391.41
  13. Wheeled humanoid robot navigation and path planning,TP242
  14. Petri net -based network intrusion detection system Research and Implementation,TP393.08
  15. Based on mid-level visual features and high-level structural information complementary target tracking model,TP391.41
  16. Stereoscopic three-dimensional reconstruction -related technology research and implementation,TP391.41
  17. Embedded target detection and tracking system design and algorithm implementation,TP391.41
  18. Target Tracking Based on Particle Filter Algorithm and DirectShow realized,TP391.41
  19. Video Mean-Shift tracking algorithm applied research,TP391.41
  20. Based on Particle Filter and reliable visual tracking technology,TP391.41
  21. Distributed Microphone Array Tracking Algorithm,TN912.3

CLC: > Industrial Technology > Automation technology,computer technology > Automation technology and equipment > Robotics > Robot
© 2012 www.DissertationTopic.Net  Mobile