Pedestrian detection and tracking using probabilistic graphical models (2016)
Pedestrian detection and tracking is one of the most challenging tasks in the fields of image sequence analysis and computer vision. Though there is a lot of work related to the detection and tracking of pedestrians, only few papers address the geometric accuracy of a detection result. For many realistic applications like motion analysis of people in sports, video surveillance and driver assistance systems, where one has to decide whether a pedestrian does actually enter a vehicle path or not, geometric accuracy is crucial.
|a) Input image||b) Prior scene information||c) Pedestrian detection|
|d) Instance specific classification||e) Motion model||f) Inferred person's position|
The aim of this project is hence to develop novel approaches for the detection and tracking of pedestrians in challenging real-world outdoor scenarios with the focus on geometric accuracy of automatically generated pedestrian annotations. Most tracking approaches use variants of the recursive Bayes filter in order to find a compromise between image based measurements (i.e. automatic pedestrian detections) and a motion model, where the motion model implies the expected temporal dynamics of the objects. In such filter models, the state variables are modeled as unknowns and the image based measurements as observations. In this context we propose a Dynamic Bayes Network in which the position of a tracked object in the image is also modeled as hidden variable. This way the system allows the detection to be corrected before it is incorporated into the recursive filter. Figure 1 shows an exemplary image taken from a video sequence and derived measurements used for the inference of the pedestrian's location. As observations prior scene information, results of a pedestrian detector, an instance specific classifier and a motion model are used.
Klinger, T.; Rottensteiner, F.; Heipke, C. (2014): Pedestrian Recognition and Localisation in Image Sequences as Bayesian Inference. In: Kukelová Z, Heller J. (Eds.), Proceedings of the Computer Vision Winter Workshop 2014 , Czech Society for Cybernetics and Informatics, S. 51-58. | file |
Klinger, T.; Rottensteiner, F.; Heipke, C. (2014): A Dynamic Bayes Network for visual Pedestrian Tracking. In: IntArchPhRS vol. XL-3, ISPRS Technical Commission III Symposium, Zurich, Switzerland, September 2014, pp. 145-150 | file |
Klinger, T.; Rottensteiner, F.; Heipke, C. (2015): Probabilistic Multi-Person Tracking using Dynamic Bayes Networks. In: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume II-3/W5, pp. 435-442, 2015 | file |