|
|
|
| Tracking a varying number of targets in videos based on
probability hypothesis density filtering |
| 1.School of Mechanical Engineering, Jiangnan University, Wuxi, Jiangsu 214122， China； 2.Jiangsu Key Laboratory of Advanced Food Manufacturing Equipment & Technology, Wuxi, Jiangsu 214122， China |
|
|
|
|
Abstract To solve the problem that probability hypothesis density (PHD) filter was not able to track birth targets of unknown positions, a PHD filtering framework was proposed based on new target detection. To overcome the inability to yield target tracks of original PHD filter, a ″trackstate estimate″ association algorithm was designed to present mathematical formulation and implementation method for track recognition. The implementation of the multitarget tracking framework for a varying number of targets was proposed. The proposed algorithm was realized by MATLAB, and two pedestrian surveillance data sets with new targets and occlusion were adopted to evaluate the performance. The results show that the proposed tracker can improve the response of PHD filter to new births and targets after occlusion by updating the intensity of new birth targets in terms of position observations. The birth targets of unknown positions can be tracked in the scenario at any time with good accuracy of target number and state estimation.
|
|
|
|
|
|
|
|
|
2015, Vol. 36 
