Nelson, R. C. (1991). Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1):33--46.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....mainly due to regions that do not fit the 2D motion assumption, since they are also segmented after the difference image is thresholded. Temporal median filters combined with velocity tuned filters are used in [9] to detect independently moving objects (IMOs) Qualitative approaches [10, 11] to the detection of IMOs could also benefit from electronic image stabilization. During these tests the system was set to use search windows and SSD windows of the same size (7 Theta 7) For these settings, the system is able to process about 20 frames per second. Table 1 shows how the frame ....

....violations of qualitative rigid world constraints. This provides a uniformly applicable strategy by which small regions of the scene can be selected for more thorough inspection. In previous work, we produced real time algorithms for detecting independently moving objects from a moving platform [10]. These techniques are more general than those based on affine 21 stabilization of the visual field, and can function in situations containing substantial motion parallax at different depths, and skewed, non planar radial flow (such as that produced in the near field during locomotion through ....

[Article contains additional citation context not shown here]

R.C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision 7:33--46, 1991.

....computational expensive correlation based approaches can provide acceptable results. One major diculty attached to optic ow approaches is the sensitivity of the ow eld for even small rotations of the camera and other mechanical disturbances. 3 In [8] the constraint ray approach described in [9] is applied for vehicle detection. October 31, 2001 DRAFT 5 In [10] and [11] feature based tracking approaches are employed for background subtraction. First the projected motion of several background points is estimated. This is used to derive the parameters of the ane motion model that ....

R. C. Nelson, \Qualitative detection of motion by a moving observer," Int. J. of Computer Vision, vol. 7, no. 1, pp. 33-46, 1991.

....of locations on the retina which have flow vectors that could not possibly originate from the rigid motion of the observer. For more advanced systems this recognition could take place on various representations of information using different detection mechanisms suited to well defined situations [Nelson, 1991]. The problem of obstacle detection could be solved by recognizing a set of locations on the retina that represent the image of a part of the 3 D world being on a collision course with the observer. To perform this task it is not necessary to compute the exact motion between the observer and any ....

R.C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7:33--46, 1991.

....at maximum, the power of higher dimensional descriptions. It is then essential to count on experimental results that support their efficiency and demonstrate their power. The importance of motion recognition problems is evidenced by the increasing attention they have received in recent years [24, 25, 1, 16, 20, 19]. One of the main areas of research is the analysis of humans in motion. There are numerous domains that motivate the research in this area: video surveillance, human computer interaction, athletics, dance, robot motion, among others. 1 Motion of objects is much more semantically rich than ....

....in which we are particularly interested. In order not to complicate the data acquisition process, we assume a static background or the possibility to separate the motion of the object from that of the camera or other objects. There are computer vision techniques that partially solve these problems [16]. Here we will concentrate more on the recognition of the actions and will simplify the registration process to an acceptable level by assuming a static background. 4.2 Motion Energy Images A MEI [6] is basically a cumulative motion image. It is a simple but probably useful representation of the ....

R. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1), 1991.

....causing apparent motion upwards in the image. Given this image triplet as input the ultimate goal would be to identify that the robot has moved independently of the background. Motion segmentation turns out to be a most demanding problem and has received considerable attention over the years (see [5, 14, 13, 22, 30, 35, 41, 45] for a review) Many previous approaches to motion clustering have failed because the motion models that they employ are inappropriate. For instance, if one tries to group based purely on similarity of image velocities then any stream of images 2 from a static scene viewed by a camera undergoing ....

R.C. Nelson. Qualitative detection of motion by a moving observer. IJCV, pages 33--46, 1991.

.... end we use the optic flow technique [11, 17, 12] Although gradient based optic flow was found in earlier structure from motion studies [18] to be too error prone for detailed quantitative recovery of scene structure, recent work on qualitative motion understanding has revived interest in its use [12, 15, 19, 20], and this is the approach we adopt. The global linear velocity field to the optic flow data is valid for a small field of view and planar surfaces, and this is adequately supported by our experimental results presented later in Section 5.The residual error of the global least squares fit turns ....

R.C.Nelson. Qualitative detection of Motion by a moving observer. Intl. Journal of Computer Vision, 7(1):33-46, 1991.

....the moving actor, and the normalization of its motion, both in the space and time dimensions, so that it can be compared to the motion templates. The first step makes sure that the system isolates the segments of the image where a moving actor exists, using information based on the optical flow [Nel91] Once the object of interest is identified, the pixels of the image that correspond to it are clustered, and tracked. The algorithm uses information from the last k image frames to estimate the centroid velocity of the object, and its future position. This makes the process robust to the ....

Randal Nelson. Qualitative detection of Motion by a Moving Observer. International Journal of Computer Vision, 7(1):33--46, November 1991.

.... are further processed by the brain to solve many visual tasks, the most fundamental ones of which include controlling eye movement [Daniilidis, 1996; Stark, 1968; Coombs and Brown, 1993] tracking [Ferm uller and Aloimonos, 1992; Smith and Brady, 1995; Koller et al. 1993] segmenting the scene [Nelson, 1991; Zeki et al. 1993] estimating 3D motion and reconstructing the scene in view or some representations of the third space dimension [Daniilidis, 1992; Faugeras, 1992] Fig. 1. A pattern similar to one by Ouchi [ 1977 ] For humans and other primates it is considered that two dimensional image ....

R. C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7:33--46, 1991.

....from the observer in his visual field. Recently, this has been approached mainly by assuming some knowledge about the observer s motion. Thomson uses knowledge of certain aspects of egomotion and scene structure [3] Sharma and Aloimonos [4] assume known translational egomotion; Nelson [5] requires a priori knowledge of egomotion parameters and assumes upper bounds on the depth of the scene. The problem of independent motion detection takes a special form in the case where the observer does not move relative to the static 3D environment. In this case, the problem of detecting ....

....parameters of the observer or of the objects in the field of view. The method is in fact a data driven, feature based attention mechanism [6] which can be exploited by a monocular observer pursuing unrestricted 3D motion. The problem of detection of maneuvering objects has also been approached in [5] by assuming smooth observer motion and inexact knowledge of the motion field. In our approach the observer s motion is not restricted and, moreover, changes in his 3D motion parameters are signaled. In addition, no assumptions about the objects motion are imposed, making it useful in practical ....

[Article contains additional citation context not shown here]

Nelson RC. Qualitative Detection of Motion by a Moving Observer. International Journal of Computer Vision, 7(1):33--46, 1991.

....known translational egomotion is hypothesized. For such an egomotion, normal flow values are restricted to lie on a half plane determined by the direction of translation. Patterns inconsistent with those defined by such half planes are considered to be due to independently moving objects. Nelson [6] presents two methods for independent motion detection which are also based on the normal flow field. The first of these methods requires a priori knowledge of egomotion parameters and assumes upper bounds on the depth of the scene. The second method detects abrupt changes of independent motion ....

....motion detection does not rely on the computation of optical flow, rather on normal flow, i.e. the projection of optical flow on the direction of the intensity gradient. The normal flow field has been used in the past for both egomotion estimation [13, 14, 15] and independent motion detection [5, 6]. In order to compute the normal flow field, a sequence of images is modeled as a continuous irradiance function I(x; y; t) of two spatial (x; y) and one temporal (t) variables. Assuming that irradiance is conserved between two consecutive frames, the well known optical flow constraint equation, ....

Nelson RC. Qualitative Detection of Motion by a Moving Observer. International Journal of Computer Vision, 7(1):33--46, 1991.

....detection. To our knowledge, this is the first work that incorporates both quantitative and qualitative motion estimates into a single algorithm. Currently, most research has been focused on how to quantitatively estimate the motion parameters [2, 3, 4] On the other hand, qualitative approaches [5, 6] have been investigated in the context where numerical motion estimates are unavailable, unstable, or unnecessary. Little attempt has been made to pursue quantitative and qualitative tracking simultaneously. Our work, however, will contribute to this direction. image range motion car tracker car ....

R.C. Nelson, "Qualitative Detection of Motion by a Moving Observer," Int. J. of Computer Vision, Vol. 7, No. 1, pp. 33-46, 1991.

.... Category Type Difficulties Parallel computers Connection machine [7] 35] 48] 51] Parsytec transputer [42] and hybrid pyramidal vision machine (AIS4000 and CSA transputer) 13] high cost, weight and power consumption Special image processing hardware PIPE [1] 11] 39] 46] Datacube [34] and PRISM 3 [36] low precision Dedicated VLSI chips Vision Chips: gradient method [31] 44] correspondence method [12] 43] and biological receptive field design [14] 30] low resolution Non Vision Chips: analog neural networks [20] digital block matching technique [4] 18] 49] coarsely ....

....is limited and radio links require considerable bandwidth and direct lineof sight. At the Perception Systems Laboratory at the National Institute of Standards and Technology, we have experienced these problems. Specialized image processing hardware such as PIPE [1] 11] 39] 46] Datacube [34] , and PRISM 3 [36] have been used for flow estimation. These machines achieve high speed with simplified arithmetic circuitry, thus the algorithms accuracies are compromised. This approach is best used in qualitative motion applications such as obstacle avoidance. Finally, there is the use of ....

Nelson, R., "Qualitative Detection of Motion by a Moving Observer", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Lahaina, HI, pp. 173-178, 1991.

....power than is practical for a mobile robot. TWR94] Authors rarely report the computational time needed for their algorithms; when they do, it is on the order of many minutes per frame ( WM93] or require specialized hardware such as a Connection Machine ( BLP89a] WZ91] L88] Datacube ([N91], LK93] or custom image processors ( DW93] One obvious reason calculating optical flow is so computationally intensive is that images are composed of thousands of pixels (which often motivates massively parallel implementations) Another reason is that optical flow is very sensitive to noise, ....

R.Nelson, Qualitative Detection of Motion by a Moving Observer, Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, p.173-178,1991

....difficulty of detecting independent motion from visual flow can be traced to the non linear nature of the problem, and the high dimensionality of visual motion information to be processed. Previous approaches typically employ simplifications, such as restricting the allowable camera motion [TP90, Nel97, SA96, CZ96, WD96] using a small camera field of view (FOV) ABP96, IA96] or imposing a limitation on the allowable structure in the scene [Adi85, MDD 95, IA96] often they employ techniques requiring point correspondence [Adi85, IA96, MDD 95, TP90, dVLT92, Smi96, WD96] or high ....

....or optimize our detection capabilities by relocating the FOE so that the moving object is placed within the region of p detectability in the image. Unfortunately, this simple and rather straightforward approach is applicable only if the motion of the camera is known or purely translational [Nel97, SA96] The problem becomes more challenging if the camera has unknown general motion [Adi85, Smi96, AO97] and the scene structure in view does not fit simple models (such as planarity, layered structure, etc. ABP96, IA96] especially if the camera FOV is large. 3 Global signatures of rigid ....

R. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 19, 1997.

.... the tracking problem has been faced in many different ways: ffl Computing the optical flow field and then parsing it in order to detect anomalies ; for instance it has been shown that optical flow due to the motion of the target is uncorrelated from optical flow due to the motion of the observer [Nelson, 1991]. ffl Computing the optical flow field and then subtracting the flow field due to the measured motion of the observer [Murray et al. 1993] As in the previous case the approach requires a considerable computational power [Heeger and Hager, 1988] and results are commonly affected by errors whose ....

R.C. Nelson, "Qualitative Detection of Motion by a Moving Observer," Int. Journal of Computer Vision, 7(1):33--46, 1991.

....mainly due to regions that do not fit the 2D motion assumption, since they are also segmented after the difference image is thresholded. Temporal median filters combined with velocity tuned filters are used in [8] to detect independently moving objects (IMOs) Qualitative approaches [9] to the detection of IMOs could also benefit from electronic image stabilization. During these tests the system was set to use search windows and SSD windows of the same size (7 Theta 7) For these settings, the system is able to process about 20 frames per second. The table in Figure 7 shows ....

R. Nelson. Qualitative Detection of Motion by a Moving Observer. International Journal of Computer Vision, 7:33-46, 1991.

....between points on different sides of flow discontinuities to discriminate between motion and depth boundaries. Techniques using as input the spatio temporal derivatives of the image intensity function (or the so called normal flow) have been developed in [Sharma and Aloimonos, 1991] and [Nelson, 1991]. For an observer moving with known translational motion Sharma and Aloimonos detect places in the image where the normal flow vector s direction is inconsistent with the range of possible directions corresponding to the actual translation. With such a strategy only a small number of ....

....is inconsistent with the range of possible directions corresponding to the actual translation. With such a strategy only a small number of discontinuities will be recognized. In order to increase this number they repeatedly change the direction of translation and perform new computations. Nelson [1991] developed two different techniques for independent motion detection. He describes the geometrical area in which a normal flow vector lies if the observer s motion is known and the depth of the scene is constrained by an upper bound. With the first technique normal flow vectors are detected as ....

R.C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7:33--46, 1991.

.... approach Category Type Difficulties Parallel computers Connection machine [6] 26] 37] 39] Parsytec transputer [32] and hybrid pyramidal vision machine (AIS 4000 and CSA transputer) 11] high cost, weight and power consumption Image processing hardware PIPE [1] 9] 29] 35] Datacube [25] and PRISM 3 [27] low precision Dedicated VLSI chips Vision Chips: gradient method [23] 34] correspondence method [10] 33] and biological receptive field design [12] 22] low resolution Non Vision Chips: analog neural networks [17] digital block matching technique [3] 15] 38] coarsely ....

Nelson, R., "Qualitative Detection of Motion by a Moving Observer", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Lahaina, HI, pp. 173-178, 1991.

.... end we use the optic ow technique [11, 17, 12] Although gradient based optic ow was found in earlier structure from motion studies [18] to be too error prone for detailed quantitative recovery of scene structure, recent work on qualitative motion understanding has revived interest in its use [12, 15, 19, 20], and this is the approach we adopt. The global linear velocity eld to the optic ow data is valid for a small eld of view and planar surfaces, and this is adequately supported by our experimental results presented later in Section 5.The residual error of the global least squares t turns out to ....

R.C.Nelson. Qualitative detection of Motion by a moving observer. Intl. Journal of Computer Vision, 7(1):33-46, 1991.

....examples from navigation: The problem of independent motion detection by a moving observer usually has been addressed with techniques for segmenting optical flow fields. But it also may be tackled through the recognition of non rigid flow fields for a moving observer partially knowing its motion [3, 41, 61]. The problem of obstacle detection could be solved by recognizing a set of locations on the retina that represent the image of a part of the 3D world being on a collision course with the observer. To perform this task it is not necessary to compute the exact motion between the observer and any ....

....We usually concentrate our attention on the moving objects in the center of the visual field (where the image is sensed with high resolution) and pay attention only if something is moving fast in the periphery. It thus seems to make sense to develop processes that detect anything moving very fast [41]. If some upper bound on the observer s motion is known (maximal speed) it is possible to detect even for small areas where motions above the speed threshold appear. Similarly, for specific systems, processes that recognize specific types of motion may be devised by employing filters that respond ....

R. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7:33--46, 1991.

....power than is practical for a mobile robot. TWR94] Authors rarely report the computational time needed for their algorithms; when they do, it is on the order of many minutes per frame ( WM93] or require specialized hardware such as a Connection Machine ( BLP89a] WZ91] L88] Datacube ([N91], LK93] or custom image processors ( DW93] One obvious reason calculating optical flow is so computationally intensive is that images are composed of thousands of pixels (which often motivates massively parallel implementations) Another reason is that optical flow is very sensitive to ....

R.Nelson,Qualitative Detection of Motion by a Moving Observer, Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, p.173-178,1991

....to support real time performance. Authors rarely report the computational time needed for their algorithms; when they do, it is on the order of many minutes per frame ( WM93] or require specialized hardware such as a Connection Machine ( BLP89a] DN93] SU87] WZ91] L88] Datacube ( LK93] [N91]) custom image processors [DW93] or PIPE [KSL85] WWB88] ATYM] Techniques which can run in real time often impose strict restrictions on the environment; HB88] presents a technique that can segment in real time for tracking purposes, but requires that a textured object be moving in front of ....

R.Nelson,Qualitative Detection of Motion by a Moving Observer, Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, p.173-178,1991

....violations of qualitative rigid world constraints. This provides a uniformly applicable strategy by which small regions of the scene can be selected for more thorough inspection. In previous work, we produced real time algorithms for detecting independently moving objects from a moving platform [ Nelson, 1991 ] These techniques are more general than those based on affine stabilization of the visual field, and can function in situations containing substantial motion paral lax at different depths, and skewed, non planar radial flow (such as that produced in the near field during locomotion through ....

....In previous work, we developed methods for detecting three qualitative types of motion. The first technique, which we term constraint ray filtering, provides a robust method of detecting independently moving objects from a moving platform when information is available about the platform motion [ Nelson, 1991 ] The method is based on the observation that the projected motion at any point on the image sphere is constrained to lie on a half line (ray) in local velocity space whose parameters depend only on the observer s motion and the location of the image point. The second method, termed animate ....

[Article contains additional citation context not shown here]

R.C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision 7:33--46, 1991.

....segmentations are produced at depth discontinuities when the observer is translating. Due to the primitive measurements used, these methods cannot cope with such confusion. b) The second group of approaches assumes that partial or full information about egomotion is available or can be recovered [1, 6, 7, 13, 20, 23 25, 29]. Most of these studies assume the availability of a full motion field; for example, Thompson and Pong suggest that if the egomotion parameters are known completely then the flow field can be checked for inconsistencies with the epipolar constraint arising from the presence of moving objects. ....

....types of independent motion, a detection scheme based solely on this constraint may miss the moving object. If under some circumstances one can put a bound on how close or how far objects may be to the observer, then more constraints can be imposed by checking the estimates against these bounds [20]. It is clear that if the egomotion is known, or assumed to be computed accurately, the detection of moving objects and estimation of their motions becomes much easier; however, perceptual systems do not treat the problem in this way. A very interesting question is how perceptual systems are able ....

R. C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7:33--46, 1991.

.... stabilized sequence is appropriate for visualization purposes, it cannot be directly applied to segment IMOs from simple difference of frames, mainly due to regions that do not fit the 2D motion assumption (see the thresholded difference of stabilized frames in Figure 2) Qualitative methods [9] and temporal median filters combined with velocity tunned filters [8] could be used to detect IMOs in such conditions. 4 CONCLUSION We presented in this paper an electronic image stabilization system based on a two dimensional feature based multi resolution motion estimation algorithm, that ....

R. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7:33--46, 1991.

....going to hit me Instead, a much simpler calculation can be performed to find the focus of expansion (FOE) and then based on the FOE location (i.e. whether it is inside or outside the image) make predictions about object motion. A catalogue of such methods are well summarized in a paper by Nelson[17]. Ian Horswill tries to explicitly identify such environmental constraints for use in navigation[10] For example, in trying to identify other independent agents, a rule such as moving objects indicate agency can be used. This rule, however, is not universal and can depend strongly on the ....

Randal C. Nelson, Qualitative detection of motion by a moving observer, International journal of computer vision, 1991

....to segment IMOs from simple difference of frames, mainly due to the clutter produced by the regions that do not fit the 2D motion assumption and small misalignments of the registration process, but observe that the moving object present in the sequence is clearly segmented. Qualitative methods [10] and temporal filters combined with velocity tuned filters [9] were suggested to detect IMOs in such conditions. 4. Conclusion We presented in this paper a fast electronic digital image stabilization system based on a two dimensional feature based multi resolution motion estimation algorithm, that ....

R. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7:33--46, 1991.

....2 : The inherent shortcomings of this equation are well explored [16] as are the shortcomings of the visual motion derived using it, particularly in terms of the difficulty in obtaining 3D structure. However, recent work on qualitative motion understanding has revived interest in its use [16, 17, 18], and certainly we find that the motion derived is sufficiently good for the 2D qualitative and quantitative interpretation we require. 2 Although this work uses only one camera, we will continue to use vergence to denote left right movements of a camera. 3 As the cameras are themselves ....

R. C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1):33--46, November 1991. 11

....approach is efficient enough to allow real time tracking. Finally, this technique also allows detection of foreground objects even when they are not moving. A moving camera can also be used, by building a background mosaic and estimating camera motion, or detecting independently moving objects [23, 34, 16]; however, such generality at the cost of a considerable increase in computation required. Another possible approach is to measure the optical flow and use it to guide registration and or segmentation. Unfortunately, non rigid body tracking cannot easily benefit from optical flow computation. This ....

R. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1), 1991.

....to support real time performance. Authors rarely report the computational time needed for their algorithms; when they do, it is on the order of many minutes per frame ( WM93] or require specialized hardware such as a Connection Machine ( BLP89a] DN93] SU87] WZ91] L88] Datacube ( LK93] [N91]) custom image processors [DW93] or PIPE [KSL85] WWB88] Techniques which can run in real time often impose strict restrictions on the environment; HB88] presents a technique that can segment in real time for tracking purposes, but requires that a textured object be moving in front of a ....

R.Nelson,Qualitative Detection of Motion by a Moving Observer, Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, p.173-178,1991

....which is derived from the light received over time on the retinae of the eyes. These image motion measurements are further processed by the brain to perform many visual tasks, the most fundamental of which include controlling eye movement [4, 39, 2] tracking [10, 38, 25] segmenting the scene [32, 46], estimating 3D motion and reconstructing the scene or some representation of the third spatial dimension [3, 9] For humans and primates it is believed that two dimensional image measurements are computed which correspond to velocity measurements of image patterns, called optical flow. The ....

R. C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7:33--46, 1991.

....at maximum, the power of higher dimensional descriptions. It is then essential to count on experimental results that support their efficiency and demonstrate their power. The importance of motion recognition problems is evidenced by the increasing attention they have received in recent years [24, 25, 1, 16, 20, 19]. One of the main areas of research is the analysis of humans in motion. There are numerous domains that motivate the research in this area: video surveillance, human computer interaction, athletics, dance, robot motion, among others. Motion of objects is much more semantically rich than static ....

....in which we are particularly interested. In order not to complicate the data acquisition process, we assume a static background or the possibility to separate the motion of the object from that of the camera or other objects. There are computer vision techniques that partially solve these problems [16]. Here we will concentrate more on the recognition of the actions and will simplify the registration process to an acceptable level by assuming a static background. 4.2 Motion Energy Images A MEI [6] is basically a cumulative motion image. It is a simple but probably useful representation of the ....

R. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1), 1991.

....regularization using Markov Random Field models. Sharma and Aloimonos [4] have proposed a method which uses the spatiotemporal derivatives of the image intensity function (normal flow field) rather than optical flow. However, as in the case of [1] translational egomotion is hypothesized. Nelson [5] presents two methods for independent motion detection which are also based on the normal flow field. The first of these methods requires a priori knowledge of egomotion parameters and assumes upper bounds on the depth of the scene. The second method detects abrupt changes of independent motion ....

R.C. Nelson. Qualitative Detection of Motion by a Moving Observer. International Journal of Computer Vision, 7(1):33--46, 1991.

....of a wider variety of visual data. Gradientbased optical flow seems to be an exemplar. Although often too error prone for detailed quantitative recovery of scene structure, recent work on qualitative motion understanding has revived enthusiasm for its use in simpler tasks (Verri Poggio 1989; Nelson 1991; Campani Verri 1992) 3.1. Motion detection and segmentation Components of the optical flow field r are derived from the spatio temporal gradients of the smoothed and subsampled image irradiance E(r; t) using the motion constraint equation (Fennema Thompson 1979; Horn Schunck 1981) E t ....

Nelson, R. C. 1991. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1): 33--46.

....in a real time system due to the fact that the motion parameters must be recovered before different components can be identified and that extensive amount of computation are usually required. The work by Nelson addresses the problem of identifying independently moving objects from a moving sensor [Nelson, 1990]. Two methods are discussed, one making use of the information about the motion of the observer, and the other using knowledge about how certain types of independently moving objects move. The implementations that run in real time on a parallel pipelined image processing system are described. The ....

R. C. Nelson. Qualitative Detection of Motion by a Moving Observer. TR--341, University of Rochester, April 1990.

No context found.

Randal. C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1):33--46, November 1991.

....or running is a sufficiently strong cue to allow strong normalization to be performed. If there are multiple actors in the scene, it is important to initially detect each actor and spatially isolate them. Fortunately, independent motion provides an exceptionally strong segmentation cue. Nelson [ Nelson, 1991 ] has demonstrated a real time method of detecting independently moving objects even in the case that the observer is itself moving. Using such a method, we can detect the pixels in an image sequence that exhibit motion independent of that of the background and segment the image frames into ....

R.C. Nelson. Qualitative detection of motion by a moving observer. In Proc. of IEEE CVPR, pages 173--178, 1991.

....velocity tuned filters (VTFs) that receive such sequences as input. As an alternative, our second approach uses a qualitative method for detecting independently moving objects based on available knowledge about the motion of the camera, which can be obtained from the computation of normal flow [8]. This approach requires a weaker form of stabilization in which only rotation effects need to be removed from the sequence. IMOs are identified by constraints on allowable values of the optical flow field for the resultant translation only image sequences. The following section describes our ....

....away from the camera. We can also detect objects moving in the range of 12 to 24 km h at 100 m by accumulating evidence from 3 consecutive processed frames. Trying to detect even faster objects would not be reliable because evidence from only 2 frames could be used. 4 Qualitative Detection Nelson [8] suggests a qualitative method of detecting independently moving objects termed constraint ray filtering. It is based on the observation that the projected motion at any point on the image sphere is constrained to lie on a half line (ray) in the local 2 D space of projected velocity, whose ....

R. Nelson. Qualitative Detection of Motion by a Moving Observer. International Journal of Computer Vision 7, 33-46, November 1991.

No context found.

R. C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1):33--46, November 1991.

....violations of qualitative rigid world constraints. This provides a uniformly applicable strategy by which small regions of the scene can be selected for more thorough inspection. In previous work, we produced real time algorithms for detecting independently moving objects from a moving platform [ Nelson, 1991 ] These techniques are more general than those based on affine stabilization of the visual field, and can function in situations containing substantial motion parallax at different depths, and skewed, non planar radial flow (such as that produced in the near field during locomotion through hilly ....

....previous work, we have developed methods for detecting three qualitative types of motion. The first technique, which we term constraint ray filtering, provides a robust method of detecting independently moving objects from a moving platform when information is available about the platform motion [ Nelson, 1991 ] The method is based on the observation that the projected motion at any point on the image sphere is constrained to lie on a half line (ray) in local velocity space whose parameters depend only on the observer motion and the location of the image point. The second method, termed animate ....

[Article contains additional citation context not shown here]

R.C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision 7:33--46, 1991.

....crossing of different activities can be tolerated as long as the regions can be separated again later. For this, it is important to initially detect each actor in the scene and spatially isolate them. Fortunately, independent motion provides an exceptionally strong segmentation cue. Nelson [21] has demonstrated a real time method of detecting independently moving objects even in the case that the observer is itself moving. Using such a method, we can detect the pixels in an image sequence that exhibit motion independent of that of the background and segment the image frames into ....

R.C. Nelson. Qualitative detection of motion by a moving observer. In Proc. of IEEE CVPR, pages 173--178, 1991.

No context found.

Nelson, R. C. (1991). Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1):33--46.

No context found.

Nelson, R. C. (1991) Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7: 33--46.

No context found.

R. C. Nelson, "Qualitative detection of motion by a moving observer," Int. Jal. of Computer Vision, vol. 7, no. 1, pp. 33-- 46, 1991.

No context found.

Nelson, R. C. (1991) Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7: 33--46.

No context found.

R. C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7:33--46, 1991.

No context found.

R.C. Nelson, "Qualitative detection of motion by a moving observer," Int. Journal of Computer Vision, vol. 7, no. 1, pp. 33-46, 1991.

No context found.

R.C. Nelson. Qualitative detection of motion by a moving observer. Technical report, 1990.

No context found.

R. C. Nelson. Qualitative detection of motion by a moving observer. International Journal of Computer Vision, 7(1):33--46, November 1991.

No context found.

Randal C. Nelson. 1991. Qualitative detection of motion by a moving observer. Intern. J. Comp. Vis., 7:33--46.

First 50 documents  Next 50

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