S. Soatto and P. Perona. Reducing "structure from motion": A general framework for dynamic vision part 1: Modeling. IEEE PAMI, 20(9), 1998.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....all the past into a state vector. This has computational efficiency advantages as well as providing a real time output. This allows the SfM estimates to be used in a closed loop control action such as navigating a robot. For a review of recursive structure from motion algorithms, consult [48] [49]. C. Small Baselines One fundamental difference between causal methods and epipolar and trifocal techniques is that only small baselines are available as the camera or objects are displaced incrementally. Thus, techniques that are sensitive in this small displacement range will exhibit numerical ....

....calibration of focal length. The RMS errors between vision and Polhemus estimates for this example were slightly better than those in the previous study, 1cm versus 1.67cm and 2.35 degrees versus 2.4 degrees) C. Independent Evaluation In an independent evaluation performed by Soatto and Perona [49], a technique similar to the proposed one demonstrated good results when compared to other variants. The similar algorithm (referred to as the Structure Integral Filter in their paper) was cast into a generic evaluation framework [48] and then compared with subspace filters [27] essential ....

S. Soatto and P. Perona. Reducing "structure from motion": A general framework for dynamic vision part 1: Implementation and experimental assessment. Pattern Analysis and Machine Intelligence, 20(9), September 1998.

....all the past into a state vector. This has computational efficiency advantages as well as providing a real time output. This allows the SfM estimates to be used in a closed loop control action such as navigating a robot. For a review of recursive structure from motion algorithms, consult [48] [49] C. Small Baselines One fundamental difference between causal methods and epipolar and trifocal techniques is that only small baselines are available as the camera or objects are displaced incrementally. Thus, techniques that are sensitive in this small displacement range will exhibit ....

....In an independent evaluation performed by Soatto and Perona [49] a technique similar to the proposed one demonstrated good results when compared to other variants. The similar algorithm (referred to as the Structure Integral Filter in their paper) was cast into a generic evaluation framework [48] and then compared with subspace filters [27] essential filters and fixation constraint methods. The system performed well in terms of accuracy, robustness and noise tolerance. The authors reported some sensitivity to initialization errors in the technique. However, the aforementioned bias ....

S. Soatto and P. Perona. Reducing "structure from motion": A general framework for dynamic vision part 1: Modeling. Pattern Analysis and Machine Intelligence, 20(9), September 1998.

....and the second is to recover motion and structure based on these correspondences. Here we deal only with the second subtask. It is well established in computer vision that if estimation is extended over multiple frames then the performance of structure from motion algorithms increases [26] [25]. In particular, algorithms perform even better if they make use of temporal coherence, relying on motion model assumptions [5] However, current methods involve highly nonlinear routines and necessitate either batch non linear least squares algorithms or recursive iterated extended Kalman ....

S. Soatto and P. Perona. Reducing "structure from motion": A general framework for dynamic vision. IEEE Trans. Pattern Analysis and Machine Intelligence, 20:933--942, 1998.

.... for known motion [16] motion from known structure [6, 9] or both structure and motion simultaneously [1, 4, 5, 7, 11, 17, 19, 29, 30] More recently, recursive schemes have been proposed for estimating motion independent of structure [25] or structure independent of motion [24] Soatto and Perona [28] have proposed a framework that unifies all geometric models for estimating structure and or motion from sequences of images. In order to achieve a fair evaluation of the geometric properties of each model it is necessary to employ the same estimation technique and the same dynamics for the ....

....structure and or motion In this section we are going to review a method for obtaining a recursive estimator of motion and or structure for all possible dynamic models derived from the constraints of rigidity and perspective. First (section 2. 1) we summarize the results of a companion paper [28], where we derive all models from the basic constraints following the idea of model reduction for dynamical systems. Then (section 2.2) we show how to transform the parameter identification task into a standard form suitable for using an Extended Kalman Filter [25] In section 2.4 we describe ....

[Article contains additional citation context not shown here]

S. Soatto and P. Perona. Reducing "structure from motion" 1: modeling. submitted to the IEEE trans. PAMI, Nov. 1995.

....have also been presented, both in closed form under the orthographic and affine projection [31, 46] and iterative for the case of full perspective projection [1, 27, 30, 41, 42] In this paper we will be dealing with causal dynamical models for multi frame processing. In the companion paper [40] we will use such models for designing local observers, such as the Extended Kalman Filter (EKF) 20] Schemes for recursive motion estimation also abound in the literature, see for instance [2, 6, 7, 9, 16, 27, 30, 37, 41] Few of them, however, can account for a varying number of features, while ....

....in the image, b) erroneous correspondence and (c) violations of the brightness constancy assumption [3] Any algorithm for reconstructing 3 D motion and or structure in real time must handle such errors in an automatic fashion. We will discuss a test for rejecting outliers in the companion paper [40]. 2.2 Limitations of the basic model The ensemble of equations (1) 5) or (2) 5) may be viewed as either a discrete time or a continuous time dynamical system that describes the evolution of point features in space, depending upon a set of parameters that encode the rigid motion constraint. In the ....

[Article contains additional citation context not shown here]

S. Soatto and P. Perona. Reducing "structure from motion" 2: experimental evaluation. IEEE trans. PAMI, submitted Nov. 1995.

.... for known motion [16] motion from known structure [6, 9] or both structure and motion simultaneously [1, 4, 5, 7, 11, 17, 19, 29, 30] More recently, recursive schemes have been proposed for estimating motion independent of structure [25] or structure independent of motion [24] Soatto and Perona [28] have proposed a framework that unifies all geometric models for estimating structure and or motion from sequences of images. In order to achieve a fair evaluation of the geometric properties of each model it is necessary to employ the same estimation technique and the same dynamics for the ....

....structure and or motion In this section we are going to review a method for obtaining a recursive estimator of motion and or structure for all possible dynamic models derived from the constraints of rigidity and perspective. First (section 2. 1) we summarize the results of a companion paper [28], where we derive all models from the basic constraints following the idea of model reduction for dynamical systems. Then (section 2.2) we show how to transform the parameter identification task into a standard form suitable for using an Extended Kalman Filter [25] In section 2.4 we describe how ....

[Article contains additional citation context not shown here]

S. Soatto and P. Perona. Reducing "structure from motion" 1: modeling. submitted to the IEEE trans. PAMI, Nov. 1995.

....how all models for estimating motion from a dynamical system fall into a special class of implicit dynamical systems with unknown parameters on a manifold. Once a model is proposed, an optimization technique needs to be employed for estimating structure and motion. We do so in a companion paper [35], where we also evaluate all methods on a common experimental ground, which highlights some caveats when reduction is performed with an output dependent change of coordinates. 1.1 Motion and structure estimation as an optimization problem Once the geometric constraints involved in the problem ....

....techniques have also been presented, both in closed form under orthographic or affine projection [26, 41] and iteratively for the case of full perspective projection [1, 23, 25, 37, 38] In this paper we will be dealing with causal dynamic models for multi frame processing. In a companion paper [35] we will use such models for designing local recursive observers, such as the Extended Kalman Filter (EKF) 17] Relatively few schemes for recursive motion estimation exist in the literature, see for instance [2, 6, 7, 9, 15, 23, 25, 32, 37] A simple counting of the dimensions will soon convince ....

[Article contains additional citation context not shown here]

S. Soatto and P. Perona. Reducing "structure from motion" 2: experimental evaluation. submitted to the IEEE trans. PAMI, Nov. 1995.

....along any state. The issue of scale normalization is addressed in appendix A. In this section we only discuss the qualitative properties of the models described in sections 2 and 5. A thorough simulation study of the performance of these methods in comparison with other schemes may be found in Soatto and Perona (1995). Simulation setup We have generated N = 20 points on a curved surface, placed 2m in front of the viewer. These points are projected onto an ideal image plane of 500 Theta 500 pixels, covering a visual angle of approximately 30 o . Gaussian noise is added to the image projections with a ....

....structure are relative to the viewer) This generates an accumulation of errors that quickly causes the estimator to diverge under the experimental conditions described in this section. Motion independent models An EKF based upon an approximate version of the model (74) has been implemented in Soatto et al. 1995), where it has been compared to an EKF based upon the structure velocity model. The motion independent model achieves lower accuracies due to the fact that the EKF approximates the noise process with a white, Gaussian additive noise process. Structure independent models A thorough evaluation of ....

[Article contains additional citation context not shown here]

Soatto, S. and Perona, P. (1995). Reducing "structure from motion" 2: experimental evaluation. submitted to the IEEE trans. PAMI, Nov. 1995.

No context found.

S. Soatto and P. Perona. Reducing "structure from motion": A general framework for dynamic vision part 1: Modeling. IEEE PAMI, 20(9), 1998.

No context found.

Soatto S. and Perona P. Reducing "structure from motion" part 2: experimental evaluation. PAMI, 20(9):943--961, 1998.

No context found.

Soatto S. and Perona P. Reducing "structure from motion" part 1: modeling. PAMI, 20(9):933-- 942, 1998.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC