Trajectory Recognition Based on Asynchronous Hidden Markov Model
Trajectory recognition of the moving objects is the basic problem of the activity analysis. To recognize the incomplete trajectory caused by the video frame loss or the occlusion, we use the asynchronous hidden Markov model (AHMM) to improve the recogniti
- PDF / 538,966 Bytes
- 11 Pages / 439.37 x 666.142 pts Page_size
- 67 Downloads / 216 Views
Abstract Trajectory recognition of the moving objects is the basic problem of the activity analysis. To recognize the incomplete trajectory caused by the video frame loss or the occlusion, we use the asynchronous hidden Markov model (AHMM) to improve the recognition accuracy. Multi-target trajectory observations are obtained using background subtraction method in which the background model is generated in HSV color space for better shadow control. To ensure the validity of the comparison between the AHMM and the hidden Markov model (HMM), the same initial parameter set is adopted in EM algorithms for each method. The hidden states of the AHMM are estimated by the E-step. Finally, the maximum likelihood of the test samples relative to all the trained models is computed, the maximum value is saved, and the corresponding model is the recognition result. Experiments indicate that the AHMM performs better than the HMM in the recognition of the incomplete trajectory.
Keywords Trajectory recognition Incomplete trajectory Asynchronous hidden Markov model
P. Qin Y. Chen (&) Key Laboratory of Advanced Process Control for Light Industry (Ministry of Education), Jiangnan University, Wuxi, Jiangsu, People’s Republic of China e-mail: [email protected] P. Qin e-mail: [email protected] Y. Chen Key Laboratory of System Control and Information Processing (Ministry of Education), Shanghai Jiaotong University, Shanghai, People’s Republic of China
Z. Wen and T. Li (eds.), Foundations of Intelligent Systems, Advances in Intelligent Systems and Computing 277, DOI: 10.1007/978-3-642-54924-3_47, Springer-Verlag Berlin Heidelberg 2014
497
498
P. Qin and Y. Chen
1 Introduction Building intelligent systems [1] in smart environment has gained significant interest in the field of computer vision. Object trajectory-based analysis and recognition is the crucial problem in the system. We restrict that the trajectories are obtained from the video cameras in the indoor environment. Motion trajectory is an important cue for the tracking and recognition, an appropriate trajectory model can well represent the object’s behaviors by analyzing the trajectories. Therefore, a lot of research work is focused on the trajectory in recent years. Chen and Chang [2] segment the object trajectory into subtrajectories based on wavelet approach, and the motion retrieval is based on the subtrajectory model. The hidden Markov model (HMM) and some extensions are efficient for representing and recognizing activities. Bashir et al. [3] segment the trajectories and represent the subtrajectories using PCA coefficients, then using the HMM to recognize trajectory. Zhang et al. [4] propose a minimum description length principle-based grammar induction algorithm to infer the meaningful rules from the event series after segmenting the trajectories into some basic motion patterns, the HMM is used in the recognition part. Nguyen et al. [5] use the hierarchical HMM to model the behaviors from human indoor trajectories and recognize the behaviors from new
Data Loading...
