Particle Filter Vehicles Tracking by Fusing Multiple Features

Yu Wang, Xiaojuan Ban, Huan Wang, Xiaorui Li, Zixuan Wang, Di Wu, Yun Yang, Sinuo LiuIEEE Access 2019 · DOI

Abstract

Real-time and accurate vehicle tracking is a fundamental requirement for Intelligent Transportation Systems (ITS), providing the data necessary for traffic condition evaluation and command dispatch. This paper proposes a Multi-Feature Fusion Particle Filter (MFFPF) designed to overcome common tracking challenges such as occlusion, illumination changes, and similar target interference. The method fuses improved color local entropy (CLE) with Scale-Invariant Feature Transform (SIFT) features, utilizing vehicle symmetry to enhance robustness. A Kalman filter is integrated to predict motion states and assist in target re-acquisition. Experimental results demonstrate that the proposed method significantly improves tracking accuracy and real-time performance in complex traffic environments.

Introduction: Challenges in Traffic Surveillance

Reliable vehicle tracking in surveillance video is often hampered by environmental complexities. Standard trackers frequently lose targets when they move through shadows, are partially hidden by other vehicles, or are surrounded by visually similar cars.

Vehicle tracking in a complex traffic environment using the proposed particle filter method.

The authors identify three primary hurdles:

  1. Illumination Variability: Sudden changes in lighting (e.g., shadows) can cause feature drift.
  2. Target Occlusion: Large vehicles or infrastructure can temporarily hide the target.
  3. Real-Time Constraints: Complex feature extraction must be optimized for live surveillance applications.

Methodology: Multi-Feature Fusion and Symmetry

The proposed MFFPF algorithm addresses these hurdles by combining complementary features and geometric priors.

Improved Color Local Entropy (CLE)

Traditional color histograms lack spatial context. By incorporating local entropy, the algorithm captures the texture and spatial distribution of pixels within the target region. To maintain real-time performance, the authors utilize a Taylor series expansion to simplify logarithmic calculations. Furthermore, the inherent symmetry of vehicles is exploited; if one side of the vehicle is occluded, the system relies on the visible side's entropy to maintain the model.

Optimized SIFT Features

While SIFT is robust to scale and rotation, it is computationally intensive. The authors optimize SIFT by weighting feature points based on their temporal persistence and using symmetry to replicate eigenvectors, which reduces computation while increasing reliability.

The mapping process from the tracked vehicle image to the geometric model used for feature extraction.

State Prediction and Target Recovery

A Kalman filter is employed to predict the vehicle's motion state, providing a search window even when the target is temporarily lost. If re-acquisition is necessary, the system combines the predicted position with license plate recognition and SIFT matching to quickly locate the correct vehicle.

Experimental Results

The MFFPF algorithm was evaluated using the Object Tracking Benchmark (OTB), the MIT Traffic Dataset, and custom surveillance sequences.

  • Benchmark Performance: On the OTB dataset, the MFFPF achieved second place in both average precision and success rate among ten state-of-the-art trackers, including KCF and CCOT.
  • Robustness to Occlusion: Qualitative tests showed the system could maintain or rapidly resume tracking even after total occlusion, thanks to the fusion of motion prediction and unique feature matching.
  • Real-Time Efficiency: The optimizations in CLE and SIFT calculations allowed the tracker to meet the requirements for real-time surveillance applications.

Conclusion

By fusing color local entropy and SIFT features with geometric symmetry and predictive filtering, the MFFPF provides a robust solution for vehicle tracking in challenging traffic scenarios. Future work will focus on improving performance in low-light nighttime conditions where feature extraction becomes significantly more difficult.

Summary updated

Abstract

Real-time and accurate vehicle tracking by Cameras and Surveillance can provide strong support for the acquisition and application of important traffic parameters, which is the basis of the traffic condition evaluation and the reasonable traffic command and dispatch. To deal with difficult problems of vehicle tracking research in a complex environments, such as occlusion, sudden illumination change, similar target interference and real-time tracking, measures are taken as follows. Firstly, the existing color local entropy particle filter tracking method is improved. The symmetry of information entropy is used to overcome the tracking failure caused by large-area occlusion. Secondly, the SIFT feature tracking method is improved to enhance real-time performance and robustness. Thirdly, two tracking methods were combined according to their characteristics, aiming at effectively improving the quasi-determination and real-time performance of vehicle tracking. Fourthly, Kalman filter was used to predict the motion state of vehicles. According to the SIFT characteristics and license plate information of vehicles, the exact position of the lost target vehicles is quickly located. It has been verified by experiments that our method has effectively improved the accuracy and real-time performance of vehicle tracking in complex situations.
BibTeX
@inproceedings{wang2019particle,
  author = {Yu Wang and Xiaojuan Ban and Huan Wang and Xiaorui Li and Zixuan Wang and Di Wu and Yun Yang and Sinuo Liu},
  title = {{Particle Filter Vehicles Tracking by Fusing Multiple Features}},
  booktitle = {IEEE Access},
  year = {2019},
  doi = {10.1109/access.2019.2941365},
}
Figure