M.Hansen, P. Anandan, K.Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proc. IEEE Image Understanding Workshop, pages457--463, 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... 18 April 2001 Accepted: 20 July 2001 camera, especially useful where a single, real camera would limit resolution or could not be used at all [8,9] Besides video compression, video coding and editing, and automatic indexing of video data, mosaicing techniques are useful for image stabilisation [10] and building high quality images with low cost imaging equipments [11] Mosaicing is based on the fact that, in some cases, two views of the same scene can be related by a non singular linear transformation of the projective plane, called homography or collineation. This happens if the camera ....

....errors during warping, and acquisition noise. To extract only relevant moving objects, we exploit temporal coherence by tracking the centroid of the moving object over the sequence. 1.2. Related Work Direct minimisation of discrepancy in pixel intensities has been widely used to align images [2,3,8,10,13,14,17]. This technique is closely related to computing a dense approximation of the 2D motion field, i.e. the apparent motion of the image brightness pattern (the optical flow) 16] Another approach to 2D motion estimation is known as feature based [18,19] which identify and match local image as ....

Hansen M, Anandan P, Data K, Wal G, Burt P. Real-time scene stabilization and mosaic construction. Proceedings of IEEE Workshop on Applications of Computer Vision 1994

....[5, 13, 12, 11] Since it is not simple to ensure a pure rotation around the optical center, such mosaics are used only in limited cases. In more general camera motions, that may include both camera translations and camera rotations, more general transformation for image alignment are used [4, 6, 10, 18, 8]. In all cases images are aligned pairwise, using a parametric transformation like an affine transformation or planar projective transformation. A reference frame is selected, and all images are aligned with this reference frame and combined to create the panoramic mosaic. Aligning all frames to ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In ARPA Image Understanding Workshop, pages 457-- 465, Monterey, California, November 1994. Morgan Kaufmann.

....when we consider HDTV or larger images, for which the large data rates demand intelligent processing. Building a good pyramid would require for HDTV and for video rate imagery. To handle the computational burden, researches have designed, built, and fielded, so called pyramid machines[7]. These machines use multiple processors in parallel to produce a pyramid at video rate for video. These pyramid machines also have the advantage of allowing parallel processing on the data of each level which is becoming important for algorithms such as image stabilization where an affine ....

M. Hansen, P. Anandan, G. Van der Wal, K. Dana, P. Burt, "Real-time scene stabilization and mosaic construction, " Proc. of the IEEE WACV, pp. 54-62, 1994.

....find a false match. Increasing the SSD window provides better discrimination between features and decreases the probability of false matches. On the other hand, the frame rate drops considerably. The last row of Table 1 shows the performance of the stabilization system presented by Hansen et al. [5]. Their system is able to stabilize images with velocities of 320 pixels per second, running at 10 frames per second. Despite the better performance figures of our system for a search window of size Sigma3, we expect that their system may be more precise in the computation of the motion ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P.J. Burt. Real-time scene stabilization and mosaic construction. In Proc. of ARPA Image Understanding Workshop, pages 457--465, Monterey, CA, November 1994, pp. 457--465.

....= x y w d d x y w x y cos sin sin cos qq qq 001 (2) where d x , d y are the translation in pixels and q is the angle of rotation. It has been shown that in image alignment, the use of transformation models with progressive complexity reduces the computation cost [4, 10]. Although for the rigid transformation there are only three parameters to be computed, two for the translation and one for the angle of rotation, a simpler model involving translation solely can be used initially for the alignment. Then, using this translation component, the angle of rotation can ....

....images are created from the original images by averaging blocks of pixels. A translation is computed for these smaller images which is then used as an initial British Machine Vision Conference 380 position to compute the translation of the original images. A detailed description can be found in [10]. It has been found that it is important to follow all local minima to the next pyramid level, especially for document images where false matches at the lower resolution levels may occur due to the repetitive nature of text lines. 2.3 Rotation Rotating an image is a time consuming operation, in ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, P. Burt, "Real-time Scene Stabilization and Mosaic Construction", IEEE Workshop on Applications of Computer Vision - Proceedings, pp.54-62, 1994.

....about the IMO s starting position in conjunction with the computed motion characteristics enables us to track it over an image sequence. Other applications of the obtained motion in 5 formation that are presented in this thesis are simultaneous image stabilization [27] 28] and mosaicing [29] [30] 1.3 Previous Approaches Recently many advances have been made in multiple motion scenarios. Active contour trackers can be used to track an object over time with a moving camera [31] 32] However this approach requires objects with a distinct contour. In general a parametric function, ....

....assumes not only correctly estimated parameters but also that the motions present can be satisfactorily modeled with the assumed affine motion model. If subsequent frames are superimposed after they have been backprojected to the same frame of reference, a mosaic of the observed scene is obtained [29] [30] When new parts of the scene become visible due to the camera motion, these are added to a scene image larger than each single frame. In this stabilized and mosaiced sequence we can now apply dynamic scene analysis tools for the SCMO case, for example difference images. On the other hand, ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P.Burt, "Realtime scene stabilization and mosaic construction," in Image Understanding Workshop, pp. 457--465, 1994.

....Objects Figure 2. Moving blobs are extracted from a video image. Detection of blobs in a video stream is performed by a process of background subtraction using a dynamically updated background model[5] from a stable video stream (either from a static camera, or stabilised by a suitable algorithm[8]) denoted I n (x) where I is the intensity of pixel x = i; j) at frame n. Firstly, each frame is smoothed with a 3 Theta 3 Gaussian filter to remove video noise. The background model Bn (x) is initialised by setting B 0 = I 0 . After this, for each frame, a binary motion mask image Mn (x) is ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proceedings of DARPA Image Understanding Workshopp, 1994.

....issue depends on what kind of motion model is selected and how image motion is detected. It has been pointed out [1,5] that human observers are more sensitive to rotational vibrations, and only camera rotation can be compensated without the necessity of recovering depth information. Hansen et al.[2] used an affine model to estimate the motion of an image sequence. The model results in large errors for rotational estimations if there are large depth variations within images and the translation of the camera is not very small. Duric et al. 5] and Yao et al. 6] dealt with this problem by ....

....are far away. It will fail in many cases where the horizon line is not very clear or just cannot be seen, or the points along the horizon line are not so far away. The second issue is critical for eliminating unwanted vibrations while preserving the smooth motion. The algorithm in Hansen et al.[2] chose a reference frame at first, and then aligned the successive frames to that frame by warping them according to the estimated inter frame motion parameters. The reconstructed mosaic image was displayed as the result of stabilization. This approach works fine if there are no significant depth ....

[Article contains additional citation context not shown here]

M. Hansen, P. Anadan, K. Dana, G. van de Wal, P. Burt, "Realtime scene stabilization and Mosaic Construction", Proc of IEEE CVPR, 1994, 54-62.

....be registered has candidate match regions projected to a common three dimensional surface using model based photogrammetric principles and a priori knowledge of sensor state variables. This model based approach has distinct advantages over the more common transformation treatments such as affine [4] or perspective in determining the transformation function since the overlap areas can be corrected for different relief distortions and will therefore yield more accurate transformation solutions. Approaches using affine transformations, or any other polynomial transformations, produce ....

M. Hansen, P. Anandan, K. Dana, G. Van der Wal and P. Burt, "Real Time Scene Stabilization and Mosaic Construction ", Proc. DARPA Image Understanding Workshop, Nov 1996, pp. 457-465.

....method work for vibrating image sequences. Notice that a commercial off the shelf camcorder with a digital stabilizing function usually distorts the perspective geometry of an image sequence since it uses 2D translation to compensate for the vibration. Existing digital stabilization algorithms [15,16] are not designed to meet the need of keeping loci straight in an EPI. In the second stage, a Fourier spectrum over a large Gaussian windowed area of an EPI is analyzed to robustly detect multiple orientations of the EPI s motion texture. This approach is different from the commonly used locus ....

....2. 1 Motion filtering Our image stabilization is a process of eliminating vibrations as if the vehicle undergoes a translation motion with constant velocity V within the time period [0,T] This is the basic difference between our motion filtering method and other image stabilization methods (e.g. [15,16]) By inserting the time variable t, Eq. 2) can be re written as ( i x i sZ s V t a t a = 3) and the average of a i (t) i=0, N(t) in frame t is ( 1 ) 1 t a t a t a t N t a s t N i i = 4) where a t V s s N(t) Z s x i N t ( ....

M. Hansen, P. Anandan, K. Dana, G. van de Wal, P. Burt, "Realtime scene stabilization and Mosaic Construction", CVPR94, 54-62

....issue depends on what kind of motion model is selected and how image motion is detected. It has been pointed out [1,5] that human observers are more sensitive to rotational vibrations, and only camera rotation can be compensated without the necessity of recovering depth information. Hansen et al.[2] used an affine model to estimate the motion of an image sequence. The model results in large errors for rotational estimations if there are large depth variations within images and the translation of the camera is not very small. Duric et al. 5] and Yao et al. 6] dealt with this problem by ....

....are far away. It will fail in many cases where the horizon line is not very clear or just cannot be seen, or the points along the horizon line are not so far away. The second issue is critical for eliminating unwanted vibrations while preserving the smooth motion. The algorithm in Hansen et al.[2] chose a reference frame at first, and then aligned the successive frames to that frame by warping them according to the estimated inter frame motion parameters. The reconstructed mosaic image was displayed as the result of stabilization. This approach works fine if there are no significant depth ....

[Article contains additional citation context not shown here]

M. Hansen, P. Anadan, K. Dana, G. van de Wal, P. Burt, "Real-time scene stabilization and Mosaic Construction", Proc of IEEE CVPR, 1994, 54-62.

....while the cumulative support matrix for which the entry qm;n tells us how much frame n overlaps with frame m when expressed in the coordinates of frame 0 (reference frame) is symmetric. 16 Researchers at Sarnoff also consider the use of subcomposites, and refer to them as tiles [43] [44]. MANN AND PICARD: FEATURELESS ESTIMATION OF PARAMETERS 1293 Fig. 10. Support matrix and mean squared registration error defined by image sequence in Fig. 9 and the estimated coordinate transformations between images. a) Entries in table. The diagonals are one since every frame is fully ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt, "Realtime scene stabilization and mosaic construction," in Proc. ARPA Image Understanding Workshop, Nov. 1994.

....for optical flow computation[28] 3 Method 3.1 Motion Segmentation and Tracking Given an image sequence I t from a moving camera, we segment regions of independent motion. The images I t are first Gaussian filtered to reduce noise, resulting in I t . The image I t is then stabilized [14] with respect to image I t , resulting in V t;t . The images V t;t and I t are differenced and thresholded to detect regions of motion, resulting in a binary motion image: M t; 1 if I t V t;t TM 0 otherwise (2) where TM is a threshold. In order to eliminate ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In DARPA IUW, pages 457--465, Monterrey, CA, Nov. 1994.

....lattice structures inherent in similarity matrices. 3.1 Motion Segmentation and Tracking Given an image sequence I t from a moving camera, we segment regions of independent motion. The images I t are first Gaussian filtered to reduce noise, resulting in I # t . The image I # t is then stabilized [12] with respect to image I # t # , resulting in V t,t # . The images V t,t # and I # t are differenced and thresholded to detect regions of motion, resulting in a binary motion image: M t, # = # 1 if # # #I # t V t,t # # # # TM 0 otherwise (2) where TM is a threshold. In order to ....

....real time system has been implemented to track and classify objects using periodicity. The system uses a dual Processor 550MHz Pentium III Xeon based PC, and runs at 15Hz with 640x240 grayscale images captured from an airborne video camera. The system uses the real time stabilization results from [12]. We will briefly discuss how the method can be efficiently implemented to run on a real time system. In computing S # , for each new frame, only a single column that corresponds to the new frame needs to be recomputed; the remaining entries can be reused (shifted) for the updated S # . ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In DARPA Image Understanding Workshop, pages 457--465, Monterrey, CA, Nov. 1994.

....GPS navigation system; orientation parameters may have somewhat less precision. However, we cannot expect the navigational parameters to be accurate enough to predict exactly where a feature of interest, such as a road, might be, but we believe that simple image to map (model) matching techniques [6, 11] can suffice to bring them into accurate correspondence. We will need to continually update the correspondences between the observed features and the map model features. Since the data is available to us in a continuous stream, the updating process can be much simpler than one of initial ....

P Anandan, P. Burt, K. Dana, M. Hansen, and G. van der Wal, "Real-time Scene Stabilization and Mosaic Construction," Proceedings of the ARPA Image Understanding Workshop, 1994, pp. 457-465.

....In the latter approach each video image is scanned for the region which best correlates to an image template. Independently, these methods have significant shortcomings. DT tracking is impossible if there is significant camera motion, unless an appropriate image stabilisation algorithm is employed [4]. It also fails if the target becomes occluded or ceases its motion. Template correlation matching generally requires that the target object s appearance remains constant. The method is generally not robust to changes in object size, orientation or even changing lighting conditions. However, the ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, P. Burt "Real-time scene stabilization and mosaic construction" Proceedings of DARPA Image Understanding Workshop 1994, 1994

....In the latter approach each video image is scanned for the region which best correlates to an image template. Independently, these methods have significant shortcomings. DT tracking is impossible if there is significant camera motion, unless an appropriate image stabilisation algorithm is employed [5]. It also fails if the target becomes occluded or ceases its motion. Template correlation matching generally requires that the target object s appearance remains constant. The method is generally not robust to changes in object size, orientation or even changing lighting conditions. However, the ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proceedings of DARPA Image UnderstandingWorkshopp, 1994.

....to detect and characterize the periodicity. 3.1 Motion Segmentation and Tracking Given an image sequence I t from a moving camera, we segment regions of independent motion. The images I t are first Gaussian filtered to reduce noise, resulting in I t . The image I t is then stabilized [5] with respect to image I t Gamma , resulting in S t;t Gamma . The images S t;t Gamma and I t are differenced and thresholded to detect regions of motion, resulting in a binary motion image: M t; Gamma = ae 1 if fi fi I t Gamma S t;t Gamma fi fi T 0 otherwise (2) where T is a ....

....A real time system has been implemented to track and classify objects using periodicity. The system uses a dual processor 400MHz Pentium II Xeon PC, and runs at 15Hz with 640x240 grayscale images captured from an airborne video camera. The system uses the real time stabilization results from [5]. 4 Examples 4.1 Person Walking on a Treadmill The first example is of a periodic motion with no (little) translational motion, a person walking on a treadmill (Figure 2) This sequence was captured using a static JVC KYF55B color camera at 640x480 30fps, deinterlaced, and scaled to 160x120. ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In DARPA Image Understanding Workshop, Monterrey, CA, Nov. 1994.

....adopted to estimate the camera motion. Several two dimensional (2D) and three dimensional (3D) stabilization schemes are described in Davis et al. 3] For 2D models, in general all the estimated affine motion parameters are compensated for, i.e. all motion is removed from the input sequence [2, 5, 8]. Stabilization in 3D is achieved by de rotating the frames, generating a translation only sequence, or a sequence containing translation and low frequency rotation (smoothed rotation) Yao et al. 12] compensates for 3D rotation by tracking multiple visual cues, like distant points and horizon ....

....performance of several stabilization systems by using them as inputs to an automatic object tracker. The better the quality of the stabilization, the better the tracker can segment and track the moving objects. Fast implementations of 2D stabilization algorithms are presented in Hansen et al. [5], Burt and Anandan [2] and Morimoto et al. 8] Hansen et al. 5] describe the implementation of an image stabilization system based on a mosaic based registration technique using pyramidal hardware (VFE 100) Burt and Anandan [2] describe a system used in the ARPA UGV Demo II program. The system ....

[Article contains additional citation context not shown here]

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P.J. Burt. Real-Time Scene Stabilization and Mosaic Construction. In Proc. of ARPA Image Understanding Workshop, pages 457-465, Monterey, CA, November 1994.

....information is needed for both computing the blur function, and performing motion compensation during temporal Kalman filtering. Unless stated otherwise, we assume that the motion is known, since the task of estimating these motions can be accurately carried out using a number of existing methods [17 19] (we demonstrate the effect of motion estimation in the second set of simulations) The aperture time is taken to be 0.8 times the frame period. In addition to the blurring, 30dB or 40dB Gaussian noise is added to all images. The AR model parameters can be estimated on either a local or global ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt, "Real-time scene stabilization and mosaic construction," in Proceedings of the Second IEEE Workshop on Applications of Computer Vision, pp. 54--62, December 1994.

....multiple images into high resolution mosaics [13] Building aerial photomosaics has long been a staple of photogrammetry, but only recently have fully automated techniques for building mosaics been developed. Most techniques still only estimate pure translations or affine transformations [4], but some recent work has dealt with the full projective case [8] Our approach is, to our knowledge, the first to combine full projective warping with near real time performance. Our techniques for automatically aligning images into photomosaics exploit the particularly simple form of the motion ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In IEEE Workshop on Applications of Computer Vision (WACV'94), pages 54--62, Sarasota, Florida, December 1994.

....frame and somehow combining their pixel intensities. Because different solutions to the problem have evolved in different research communities, with different applications in mind, the resulting representations have received diverse names. Among these are salient stills [19, video mosaics [6, 16, 11, 18, 12], video sprites [13, and video layers [20, 5] 1 . 1 While, strictly speaking, layering allways includes the construction In spite of this diversity, all these procedures are similar in the sense that they follow the following two fundamental steps. 1. Fitting a global motion model to the ....

M. Hansen, P. Anandan, K. Dana, G. Wal, and P. Burt. RealTime Scene Stabilization and Mosaic Construction. In Proc. ARPA Image Understanding Workshop, 1994.

....moving in the video, the proposed method successfully registers all of the images in the orbit, mapping them into a single high resolution image composite of the entire playing field. Figure 12(a) 18 Researchers at Sarnoff also consider the use of sub composites, and refer to them as tiles [43][44] (a) b) Figure 12: Image composite made from 16 video frames taken from a television broadcast sporting event. Note the Edgertonian appearance, as each player traces out a stroboscopic like path. The proposed method works robustly, despite the movement of players on the field. a) Images are ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt, "Real-time scene stabilization and mosaic construction," ARPA image understanding workshop, 10 Nov 1994.

....[8, 16, 15, 14] But the limitations to motion which is a pure rotation about the optical center limits the applicability of this approach. In more general camera motions, that may include both camera translations and small camera rotations, more general transformation for image alignment are used [6, 9, 13, 21, 11]. In all cases images are aligned pairwise, using a parametric transformation like an affine transformation or planarprojective transformation. A reference frame is selected, and all images are aligned with this reference frame and combined to create the panoramic mosaic. Aligning all frames to a ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P.J. Burt. Real-time scene stabilization and mosaic construction. In ARPA Image Understanding Workshop [1], pages 457--465.

....to the forward motion of the vehicle. The system offers the user the option to compensate of not for this and the other motion parameters of the model, what helps visualization in general. A different fast stabilization system based on a mosaic based image registration technique is presented in [5]. This system uses a multiresolution, iterative process that estimates affine motion parameters between levels of Laplacian pyramid images. From coarse to fine levels, the optical flow of local patches of the image is computed using a cross correlation scheme. The motion parameters are then ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proc. DARPA Image Understanding Workshop, pages 457--465, Monterey, CA, November 1994.

.... filtering of the high frequency or oscillatory motion due to irregularities of the terrain from the desired smooth motion that would result if the vehicle were running over smooth terrain [4, 12] Fast implementations of electronic image stabilization algorithms are presented in Hansen et al. [5], and Morimoto et al. 9] Hansen et al. 5] describe the implementation of an image stabilization system based 0 The support of ARPA under contract DAAH 0493G049 and of the Conselho Nacional de Desenvolvimento Cient ifico e Tecnol ogico (CNPq) is gratefully acknowledged on a mosaic based ....

.... motion due to irregularities of the terrain from the desired smooth motion that would result if the vehicle were running over smooth terrain [4, 12] Fast implementations of electronic image stabilization algorithms are presented in Hansen et al. 5] and Morimoto et al. 9] Hansen et al. [5] describe the implementation of an image stabilization system based 0 The support of ARPA under contract DAAH 0493G049 and of the Conselho Nacional de Desenvolvimento Cient ifico e Tecnol ogico (CNPq) is gratefully acknowledged on a mosaic based registration technique using a pyramidal ....

[Article contains additional citation context not shown here]

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proc. DARPA Image Understanding Workshop, pages 457--465, Monterey, CA, November 1994.

....[5, 13, 12, 11] Since it is not simple to ensure a pure rotation around the optical center, such mosaics are used only in limited cases. In more general camera motions, that may include both camera translations and camera rotations, more general transformation for image alignment are used [4, 6, 10, 18, 8]. In all cases images are aligned pairwise, using a parametric transformation like an affine transformation or planar projective transformation. A reference frame is selected, and all images are aligned with this reference frame and combined to create the panoramic mosaic. Aligning all frames to a ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In ARPA Image Understanding Workshop, pages 457-- 465, Monterey, California, November 1994. Morgan Kaufmann.

....frame and somehow combining their pixel intensities. Because different solutions to the problem have evolved in different research communities, with different applications in mind, the resulting representations have received diverse names. Among these are salient stills [23, 15] video mosaics [8, 20, 13, 22, 14], video sprites [16, 6] and video layers [24, 5] 1 . In spite of this diversity, all these procedures are similar in the sense that they follow the following two fundamental steps. 1. Fitting a global motion model to the motion between each pair of successive frames. 2. Computing the ....

M. Hansen, P. Anandan, K. Dana, G. Wal, and P. Burt. RealTime Scene Stabilization and Mosaic Construction. In Proc. ARPA Image Understanding Workshop, 1994.

....model can be simplified by the common assumption that the scene in view is well approximated by a plane. Past approaches calculate the background motion model by matching and tracking a set of detected features in successive frames [1] using local image correlation to approximate optic flow [3], or with multi scale gradient based methods [5] This background motion model serves to stabilize the image of the background plane. Then, independent motion is detected as either residual flow [1] background subtraction, or temporal differencing of intensity [7] The approach presented here ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proc. IEEE Image Understanding Workshop, pages 457--463, 1994.

....interest in image mosaics that provide panoramic views of the scene constructed from multiple images. The mosaic image is constructed by aligning a number of 2D views of the same scene to each other and assembling them into a single image. For a detailed example of mosaic construction, see [Hansen94]. The interest in mosaics is partly motivated by applications of the mosaic representations to a number of traditional problems in visualization and video exploitation and compression [Irani94b] These current techniques, however, rely on 2D planar surface registration which is adequate in ....

Hansen, M., Anandan, P., Dana, K., van der Wal, G., and Burt, P., "Real-time scene stabilization and mosaic construction", in these proceedings.

....continually adjust the turret to keep it locked on a stationary or moving object. Additionally, the video is continuously moving, reflecting the selfmotion of the camera. The combination of these factors often leads to operator confusion and nausea. Sarnoff has built image alignment techniques [4, 12] to stabilize the view from the camera turret and to automate camera control, thereby significantly reducing the strain on the operator. In particular, real time image alignment is used to keep the camera locked on a stationary or moving point in the scene, and to aim the camera at a known ....

....overcome this problem by providing extended views of regions swept over by the camera. Figure 35c displays an aerial mosaic of the Demo I Bushy Run site. The video sequence was obtained by flying over the demo site while panning the camera turret back and forth and keeping the camera tilt constant[12, 24, 23]. The VSAM IFD team also demonstrated coarse registration of this mosaic with a USGS orthophoto using a projective warp to determine an approximate mapping from mosaic pixels to geographic coordinates. It is feasible that this technology could lead to automated methods for updating existing ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proc. Workshop on Applications of Computer Vision, 1994.

.... technology developed by Sarnoff for real time video processing, and incorporates the processing functions found in Sarnoff s earlier PYR 1 and PYR 2 pyramid processing ASICs [2] as well as numerous other functions found in Sarnoff developed video processing systems, including the VFE100 and VFE200 [3,4,5,6]. The Acadia is designed to support real time affine motion analysis and 3D recovery from stereo for demanding vision tasks such as accurate video stabilization, mosaicking, and motion stereo capabilities for vehicle navigation. In addition the chip supports real time image fusion of ....

....consisting of multiple processing modules connected to a crosspoint switch, multi port access to shared memory, flexible digital video interfaces, link ports for expansion, and a control interface. The architecture is based on pipelined pyramid processing systems previously implemented by Sarnoff [3,4,5,6]. The Acadia vision processor CAMP2000 final 2 To be published in IEEE proceedings of International Workshop on Computer Architecture for Machine Perception, Padua, Italy, September 2000. Fig.2 Acadia Chip Diagram 2.1. Pipelined processing modules The central part of the Acadia is a set of ....

[Article contains additional citation context not shown here]

M.W. Hansen, P. Anandan, K. Dana, G.S. van der Wal, and P.J. Burt, "Real-time scene stabilization and mosaic construction". Proceedings of the 2 nd IEEE Workshop on Applications of Computer Vision. December 5-7, 1994, pp. 54-62.

....or from two entirely different sensors, possibly of different modality (e.g. IR and visible) Often, the displacement field can be described as a global parametric transformation between the two images, such as an affine, quadratic, or a projective transformation. Many methods have been developed [3, 10, 7, 4, 13, 17, 16, 6, 9, 15] for the parametric registration of a pair of images. The flow based techniques (e.g. 17, 7, 13, 9] divide the processing into two stages: first a flow field is estimated, which is then followed by regression to find the global parametric transformation which best describes the flow field. ....

....can be described as a global parametric transformation between the two images, such as an affine, quadratic, or a projective transformation. Many methods have been developed [3, 10, 7, 4, 13, 17, 16, 6, 9, 15] for the parametric registration of a pair of images. The flow based techniques (e.g. [17, 7, 13, 9]) divide the processing into two stages: first a flow field is estimated, which is then followed by regression to find the global parametric transformation which best describes the flow field. However, often the local flow estimates are noisy and unreliable, resulting in poor registration accuracy ....

[Article contains additional citation context not shown here]

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proc. of the Workshop on Applications of Computer Vision II, Sarasota, Fl., 1994.

....example, normalizing the images to remove global changes in mean and contrast often handles effects of overall lighting changes. More local variations can be handled 8 by using Laplacian pyramid representations and by applying local contrast normalizations to the Laplacian filtered images (see [6] for a real time direct affine estimation algorithm which uses Laplacian pyramid images together with some local contrast normalization) A second (and more recent) approach to handling brightness variation is to generalize the entire approach to use other local match measures besides the ....

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proc. of the Workshop on Applications of Computer Vision II, Sarasota, Fl., 1994.

No context found.

M. Hansen, P. Anandan, G. van der Wal, K. Dana, P. Burt. , "Real-time scene stabilization and mosaic construction," IEEE Workshop on Applications of Computer Vision, 1994.

....FLIR Systems Ultra 3000 turret that has two degrees of freedom (pan tilt) a GPS system for measuring position, and an AHRS (Attitude Heading Reference System) device for measuring orientation. Video processing is performed using on board Sarnoff VFE 100 and PVT 200 video processing engines [ Hansen et al. 1994 ] in Years 1 and 2, respectively. Because the cost of testing algorithms on the airplane is high, we have constructed a simulated airborne platform using a gantry, shown in Figure 3. The gantry travels in the XY plane with a camera suspended beneath it on a pan tilt mount. This enables us to ....

....turret to keep it locked on a stationary or moving target. Additionally, the video is continuously moving, reflecting the ego motion of the camera. The combination of these factors often leads to operator confusion and nausea. We have built upon image alignment techniques [ Bergen et al. 1992, Hansen et al. 1994 ] to stabilize the view from the camera turret and used the same techniques to automate the camera control, thereby significantly reducing the strain on the operator. In particular, we use real time image alignment to keep the camera locked on a stationary or moving point in the scene, and to aim ....

[Article contains additional citation context not shown here]

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Realtime scene stabilization and mosaic construction. In Proc. Workshop on Applications of Computer Vision, 1994.

No context found.

M.Hansen, P. Anandan, K.Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In Proc. IEEE Image Understanding Workshop, pages457--463, 1994.

No context found.

M. Hansen, P. Anandan, K. Dana, G. van der Wall, and P. Burt. Real-time scene stabilization and mosaic construction. In Proceedings of the Second IEEE Workshop on Applications in Computer Vision, pages 54--62, 1994.

No context found.

Hansen, M.; Anandan, P.; Dana, K.; van der Waal, G.; Burt, P. Real-time scene stabilization and mosaic construction. Proceedings IEEE Workshop on Applications of Computer Vision; 1994; pp.54-62.

No context found.

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt. Real-time scene stabilization and mosaic construction. In IEEE Workshop on Applications of Computer Vision (WACV'94), pages 54--62, Sarasota, Florida, December 1994.

No context found.

M. Hansen, P. Anandan, K. Dana, G. van der Wal, and P. Burt, \Real-time scene stabilization and mosaic construction," in DARPA Image Understanding Workshop, (Monterrey, CA), Nov. 1994.

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