Citation Context

...y the first raw estimation are now correctly classified. At this point we have a good estimate of the region of the palm and we can compute the best plane fitting the points in P using SVD and RANSAC =-=[6]-=-. Exploiting this information we get the normal of the plane that represents a good estimate of the hand orientation, a very useful information for any gesture recognition and pose estimation approach...

Citation Context

... also from H. D. Fingers Identification We based our finger detection scheme on the notion of density, therefore we decided to use Density-Based Spatial Clustering of Applications with Noise (DBSCAN) =-=[5]-=- to a) b) c) Fig. 4. Fingers and palm classification according to the ratio of the density before and after the contraction (best viewed in colors, red samples are associated to fingers, black to the ...

Citation Context

...is step we are interested in removing only artifacts, therefore the threshold to discriminate outliers is relatively high in order to reduce the chance of removing relevant points. A bilateral filter =-=[15]-=- is then applied in order to reduce the noise while preserving the details and edges of the hand shape. After these steps, a filtered point cloud H = {p1, ..., pn} containing the hand samples is avail...

Citation Context

...of degrees of freedom associated to the hand movements and to the large number of inter-occlusions between the various fingers. Several computer vision approaches have been presented for this problem =-=[4]-=-, but its solution from a single video stream remains very difficult. The recent introduction of consumer depth cameras [1] have made depth acquisition available to the mass market and has paved the w...

Citation Context

... the exploitation of 3D information acquired with consumer depth cameras. In particular some recent works focused on the estimation of the hand pose from a single depth map. The work of Keskin et Al. =-=[7]-=- has adapted the Kinect skeletal tracking scheme to hand pose estimation. A coarse-to-fine search based on Latent Regression Forests, has been proposed in [12]. Regression Forests are also used in [14...

Citation Context

...y optimization technique. Dense feature extraction exploiting convolutional networks has been used for hand pose estimation in [16]. Another approach exploiting convolutional networks is [8]. Finally =-=[13]-=- uses transductive regression forest exploiting both real and synthetic data for the training. III. PROPOSED APPROACH The target of the proposed algorithm is to identify the five fingers and the palm ...

Citation Context

...s simple in the case of raised fingers but becomes very challenging when the fingers are bent over the palm. Notice how many approaches based on the hand silhouette and also on depth data, e.g., [3], =-=[11]-=-, are able to recognize only the raised fingers and typically assign to the palm region all the samples in the lower hand area including the ones corresponding to bent fingers. After the NGC, we intui...

Citation Context

...orests, has been proposed in [12]. Regression Forests are also used in [14], where the error between the data and the model based on subdivision surfaces is optimized. Another very recent approach is =-=[10]-=- where the hand is modeled with a set of spheres and a cost function is optimized using Particle Swarm Optimization. The approach of [9] searches the optimal solution in the highdimensional parameter ...

Citation Context

...imensional parameter space of hand configurations by using an evolutionary optimization technique. Dense feature extraction exploiting convolutional networks has been used for hand pose estimation in =-=[16]-=-. Another approach exploiting convolutional networks is [8]. Finally [13] uses transductive regression forest exploiting both real and synthetic data for the training. III. PROPOSED APPROACH The targe...

Citation Context

...ngle depth map. The work of Keskin et Al. [7] has adapted the Kinect skeletal tracking scheme to hand pose estimation. A coarse-to-fine search based on Latent Regression Forests, has been proposed in =-=[12]-=-. Regression Forests are also used in [14], where the error between the data and the model based on subdivision surfaces is optimized. Another very recent approach is [10] where the hand is modeled wi...

Citation Context

...[7] has adapted the Kinect skeletal tracking scheme to hand pose estimation. A coarse-to-fine search based on Latent Regression Forests, has been proposed in [12]. Regression Forests are also used in =-=[14]-=-, where the error between the data and the model based on subdivision surfaces is optimized. Another very recent approach is [10] where the hand is modeled with a set of spheres and a cost function is...

Citation Context

...ers. Several computer vision approaches have been presented for this problem [4], but its solution from a single video stream remains very difficult. The recent introduction of consumer depth cameras =-=[1]-=- have made depth acquisition available to the mass market and has paved the way to the introduction of a set of new hand pose estimation approaches exploiting the three-dimensional information contain...

Citation Context

...d of the hand acquired by a depth sensor. Firstly the hand point cloud Hs that will be used as input for the following stages is extracted from the rest of the scene using the method proposed in [2], =-=[3]-=-. This method exploits a combination of thresholding of the depth values and geometrical constraints on the object size to recognize and segment the hand from the rest of the scene. The proposed appro...

Citation Context

...ivision surfaces is optimized. Another very recent approach is [10] where the hand is modeled with a set of spheres and a cost function is optimized using Particle Swarm Optimization. The approach of =-=[9]-=- searches the optimal solution in the highdimensional parameter space of hand configurations by using an evolutionary optimization technique. Dense feature extraction exploiting convolutional networks...

Citation Context

... cloud of the hand acquired by a depth sensor. Firstly the hand point cloud Hs that will be used as input for the following stages is extracted from the rest of the scene using the method proposed in =-=[2]-=-, [3]. This method exploits a combination of thresholding of the depth values and geometrical constraints on the object size to recognize and segment the hand from the rest of the scene. The proposed ...

Citation Context

...n evolutionary optimization technique. Dense feature extraction exploiting convolutional networks has been used for hand pose estimation in [16]. Another approach exploiting convolutional networks is =-=[8]-=-. Finally [13] uses transductive regression forest exploiting both real and synthetic data for the training. III. PROPOSED APPROACH The target of the proposed algorithm is to identify the five fingers...