## Tracking Cars in Range Images Using the Condensation Algorithm (1999)

Venue: | Proc. 1999 IEEE/IEEJ/JSAI Intl. Conf. on Intelligent Transportation Systems |

Citations: | 6 - 0 self |

### BibTeX

@INPROCEEDINGS{Meier99trackingcars,

author = {Esther B. Meier and Frank Ade},

title = {Tracking Cars in Range Images Using the Condensation Algorithm},

booktitle = {Proc. 1999 IEEE/IEEJ/JSAI Intl. Conf. on Intelligent Transportation Systems},

year = {1999},

pages = {129--134}

}

### OpenURL

### Abstract

The detection of objects in every frame of a sequence is often not sufficient for scene interpretation. Tracking can increase the robustness, especially when occlusions occur or when objects temporarily disappear. In this paper we present a stochastic tracking approach which is based on the CONDENSATION algorithm -- conditional density propagation over time -- that is capable of tracking multiple objects with multiple hypotheses in range images. A probability density function describing the likely state of the objects is propagated over time using a dynamic model. The measurements influence the probability function and allow the incorporation of new objects into the tracking scheme. Additionally, the representation of the density function with a fixed number of samples ensures a constant running time per iteration step. Results with data from different sources are shown for automotive applications. 1.

### Citations

1216 | CONDENSATION: conditional density propagation for visual tracking
- Isard, Blake
- 1998
(Show Context)
Citation Context ... not sufficient to establish a safe control system. To increase the robustness, a tracking process which can handle hypotheses has to be included. Our work has evolved from the CONDENSATION algorithm =-=[7, 8]-=- developed for contour tracking in visual clutter. Outlines and features of moving foreground objects, modeled as curves, are tracked in video sequences. Some elements in the background clutter may co... |

842 |
Applied Optimal Estimation
- Gelb
- 1974
(Show Context)
Citation Context ...solved by hypotheses tracking where probabilistic models of object shape and motion are applied to analyze the video-stream. We propose to apply the CONDENSATION algorithm instead of Kalman filtering =-=[3, 6]-=- which has been discussed thoroughly in the literature. A feature-based approach is presented in [2] where points or lines are updated with a Kalman filter. Problems arise in the subsequent grouping o... |

587 | Contour tracking by stochastic propagation of conditional density
- Isard, Blake
- 1996
(Show Context)
Citation Context ... not sufficient to establish a safe control system. To increase the robustness, a tracking process which can handle hypotheses has to be included. Our work has evolved from the CONDENSATION algorithm =-=[7, 8]-=- developed for contour tracking in visual clutter. Outlines and features of moving foreground objects, modeled as curves, are tracked in video sequences. Some elements in the background clutter may co... |

199 | Robust multiple car tracking with occlusion reasoning
- Koller, Weber, et al.
- 1994
(Show Context)
Citation Context ...f this approach is that even in the presence of partial occlusion, some features of the objects usually remain visible. Tracking based on object models is comparable to the feature-based approach. In =-=[10, 4]-=- Kalman filters are applied to estimate bounding contours of moving objects, where the contour extraction is based on motion and grey value boundaries. In highway scenes with heavy traffic, the proble... |

182 | Model-based object tracking in monocular image sequences of road traffic scenes - Koller, Daniilidis, et al. - 1993 |

110 | A real-time computer vision system for measure traffic parameters
- Beymer
- 1997
(Show Context)
Citation Context ...yze the video-stream. We propose to apply the CONDENSATION algorithm instead of Kalman filtering [3, 6] which has been discussed thoroughly in the literature. A feature-based approach is presented in =-=[2]-=- where points or lines are updated with a Kalman filter. Problems arise in the subsequent grouping of image features in this tracking process. Common motion constraints can be used to group the object... |

30 |
A Pattern Theoretical Study of Biological Shapes
- HANDS
- 1991
(Show Context)
Citation Context ...the space on which x is defined is multi-dimensional and large, we can not sample p(x t jZ t ) at regular intervals. Therefore, we must use iterative sampling techniques. The factored sampling method =-=-=-[5] is used to find an approximation to a probability density f(x) = f 2 (x) f 1 (x); x 2 X: (4) A set of samples fs (1) ; ; s (N) g with s (n) 2 X is drawn randomly from f 1 (x). By choosing a sam... |

30 | Qualitative and quantitative car tracking from a range image sequence
- Zhao, Thorpe
- 1998
(Show Context)
Citation Context ... that the generic models can handle all kinds of cars that can be found on motorways. In addition dynamic models, which describe the motion of an object, can also be incorporated into a Kalman filter =-=[16, 12]-=-. In the case of multiple hypotheses tracking caused by occlusion or temporarily disappeared objects Kalman filters can rapidly lead to unwieldy levels of complexity. Hence a mechanism has to be imple... |

23 |
Performance Assessment of Model-based Tracking
- Baker, Sullivan
- 1992
(Show Context)
Citation Context ...s shift in the object trajectory an occlusion detection can be performed by a depth ordered detection of overlapping contours. A more complex approach using 3D model-based tracking is investigated in =-=[9, 1]-=-, where a parameterized 3D generic model is used to represent various types of cars. Admittedly it seems a bit risky to expect that the generic models can handle all kinds of cars that can be found on... |

22 |
Toward Augmented Control Systems
- Pentland, Liu
- 1995
(Show Context)
Citation Context ... that the generic models can handle all kinds of cars that can be found on motorways. In addition dynamic models, which describe the motion of an object, can also be incorporated into a Kalman filter =-=[16, 12]-=-. In the case of multiple hypotheses tracking caused by occlusion or temporarily disappeared objects Kalman filters can rapidly lead to unwieldy levels of complexity. Hence a mechanism has to be imple... |

19 | Using the condensation algorithm to implement tracking for mobile robots
- Meier, Ade
- 1999
(Show Context)
Citation Context ...n exact localizations impossible. Several hypotheses of an object represent this uncertainty. In this experiment N = 60 samples were used at each time step. More examples on real data can be found in =-=[11]-=-. 7. Conclusion A stochastic tracking approach has been presented, that can handle multiple hypotheses and newly appearing objects. A probability density function is used to describe the likely state ... |

14 |
The lock-in CCD - twodimensional synchronous detection of light
- Spirig, Seitz, et al.
- 1995
(Show Context)
Citation Context ...e fixed number of hypotheses reduce the computational complexity and lead to a constant running time per iteration step. 6. Results In the project MINORA a universal miniaturized optical range camera =-=[15, 14]-=- is being developed. Since this range sensor is still under development, range image sequences of real traffic scenes are not yet available. Therefore, we use simulated range image sequences and scale... |

13 | Combining multiple motion estimates for vehicle tracking
- Gil, Milanese, et al.
- 1996
(Show Context)
Citation Context ...f this approach is that even in the presence of partial occlusion, some features of the objects usually remain visible. Tracking based on object models is comparable to the feature-based approach. In =-=[10, 4]-=- Kalman filters are applied to estimate bounding contours of moving objects, where the contour extraction is based on motion and grey value boundaries. In highway scenes with heavy traffic, the proble... |

8 |
The multitap lock-in CCD with offset subtraction
- Spirig, Marley, et al.
- 1997
(Show Context)
Citation Context ...e fixed number of hypotheses reduce the computational complexity and lead to a constant running time per iteration step. 6. Results In the project MINORA a universal miniaturized optical range camera =-=[15, 14]-=- is being developed. Since this range sensor is still under development, range image sequences of real traffic scenes are not yet available. Therefore, we use simulated range image sequences and scale... |

4 |
Modeling and Planning for Sensor Based Intelligent Robot Systems, chapter Toward Smarter Cars
- Sobottka, Meier, et al.
- 1999
(Show Context)
Citation Context ...mporarily disappeared objects Kalman filters can rapidly lead to unwieldy levels of complexity. Hence a mechanism has to be implemented for Kalman filtering to control the evolution of hypotheses. In =-=[13]-=- we present such an aging approach where hypotheses survive until their age exceeds a predefined threshold. In this paper we first introduce the mathematics needed to formulate the CONDENSATION algori... |