Daehwan Kim

Simultaneous gesture segmentation and recognition based on forward spotting accumulative HMMs

In this paper, we propose a forward spotting scheme that executes gesture segmentation and recognition simultaneously by detecting start point. By using competitive differential observation probability, sliding window and accumulative HMMs, we apply the proposed method to recognize the upper-body gestures for controlling the curtains and lights in a smart home environment

A Fast ICP Algorithm for 3-D Human Body Motion Tracking

Iterative closest point (ICP) algorithm has been widely used for registering the geometry, shape and color of the 3-D meshes. However, ICP requires a long computation time to find the corresponding closest points between the model points and the data points. To overcome this problem, we propose a fast ICP algorithm that consists of two acceleration techniques: hierarchical model point selection (HMPS) and logarithmic data point search (LDPS). HMPS accelerates the search by reducing the search region of the data points corresponding to a model point effectively: it selects the model points in a coarse-to-fine manner and employs the four neighboring closest data points in the upper layer to make the search region for finding the closest data point corresponding to a model point in the lower layer. LDPS accelerates the search by visiting the data points within the search region using 2-D logarithm search. The HMPS method and the LDPS method can be operating separately or together. To evaluate the speed of the proposed ICP, we apply it to the 3-D human body motion tracking. The proposed fast ICP is about 3.17 times faster than the existing ICP such as the K-D tree.

Real-time automatic vehicle management system using vehicle tracking and car plate number identification

This paper proposes a real-time vehicle management system using a vehicle tracking and a car plate number identification technique. The system uses two cameras: one for tracking vehicles and another for capturing LP (license plate). We track the vehicles by applying the CONDENSATION algorithm over the vehicles movement image captured from the first camera. To render the CONDENSATION algorithm more effective, we build a discrete vehicle shape model by training vehicle patterns with a SOM (self organizing map), which makes the system suitable for real-time application. Next, we take the probabilistic dynamic model such as HMM (hidden Markov model) to reflect the temporal change in shape of various vehicles. As a vehicle reaches the designated target line, a signal is sent to the second camera for capturing the vehicles front side. The captured image is transferred to an LPR (vehicle LP recognition system) which recognizes the vehicles category and LP. LPR system detects the vehicle LP using the only the vertical edge of the captured vehicle image, and effectively accomplishes the character segmentation of the LP region using the geometric transformation without respect to the position and angle of the CCD camera. The segmented characters are recognized using the SVM (support vector machine). By combining these two techniques, we construct a real-time automatic vehicle management system that can be used to control vehicle parking and searching for specific vehicles.

An Intelligent Smart Home Control Using Body Gestures

This paper proposes the control of smart home environments such as lights and curtains using body gestures. We use a forward spotting scheme that executes gesture segmentation and recognition simultaneously. The start and end points of gestures are determined by zero crossing from negative to positive (or from positive to negative) of a competitive differential observation probability that is defined by the difference of observation probability between the maximal gesture and the non-gesture. We also use the sliding window and accumulative HMMs. We apply the proposed simultaneous gesture segmentation and recognition method to recognize the upperbody gestures for controlling the curtains and lights in a smart home environment. Experimental results show that the proposed method has a good recognition rate of 95.42% for continuously changing gestures.

Pose robust human detection using multiple oriented 2d elliptical filters

This paper proposes a pose robust human detection method from a sequence of stereo images using the multiple oriented 2D elliptical filters (MO2DEFs), which can detect the humans regardless of the their scales and poses. Existing object oriented scale adaptive filter (OOSAF) has some disadvantages since they cannot detect the human with an arbitrary pose. To overcome this limitation, we introduce the pose robust MO2DEFs whose shapes are the oriented ellipses. We perform human detection by applying four 2D elliptical filters with specific orientations to the 2D spatial-depth histogram and by taking the thresholds over the filtered histograms. In addition, we determine the human pose by taking the orientation of the 2D elliptical filter whose convolution result is maximal among the MO2DEFs. We verify the human candidates by either detecting the face or matching head-shoulder shapes over the segmented human candidates of the selected rotation. The experimental results show that (1) the accuracy of pose angle estimation is about 88%, (2) the human detection using the proposed MO2DEFs outperforms that of using the existing OOSAF by 15~20%, especially in case of the posed human.

Dynamic markov random field model for visual tracking

We propose a new dynamic Markov random field (DMRF) model to track a heavily occluded object. The DMRF model is a bidirectional graph which consists of three random variables: hidden, observation, and validity. It temporally prunes invalid nodes and links edges among valid nodes by verifying validities of all nodes. In order to apply the proposed DMRF model to the object tracking framework, we use an image block lattice model exactly correspond to nodes and edges in the DMRF model and utilize the mean-shift belief propagation (MSBP). The proposed object tracking method using the DMRF surprisingly tracks a heavily occluded object even if the occluded region is more than 70~80%. Experimental results show that the proposed tracking method gives good tracking performance even on various tracking image sequences(ex. human and face) with heavy occlusion.

Self-occlusion handling for human body motion tracking from 3D ToF image sequence

A 3D Time-of-flight (ToF) image is very useful to accurately track the human body motion due to its precision. However, the ToF image can not provide occluded 3D data because it also has a limitation of camera viewpoint. This paper proposes a self-occlusion handling scheme for human body motion tracking from 3D ToF image sequence. The proposed self-occlusion handling scheme consists of two steps: detect whether the body part is occluded or not and then estimate its motion from estimating the motion of non-occluded its adjacent body parts. Occlusion can be easily detected by using the eigenvalue analysis of 3D ToF data gathered from the joint point of each body part, and their motions can be estimated by calculating the rotation of the occluded body part. To apply it to the human body motion tracking, we use the Iterative closest point (ICP) algorithm and particle filter to track even the motion of fast moving body parts. Experimental results show that the human body motion tracking with the proposed self-occlusion handling scheme can correctly estimate even the motion of the self-occluded body part by comparing the estimated joint points with the manually marked joint points.

Intuitive Pointing Position Estimation for Large Scale Display Interaction in Top-View Depth Images

In this paper, we propose an intuitive pointing position estimation method for large scale display interaction in top-view depth images. The depth sensor is mounted above the users’ head in order to avoid the sensor occluding the display. In order to estimate the pointing position, we detect the user’s head and estimate the position of the user’s eye. To calculate the center of the head, we propose a head segmentation method. We use an iterative binary partitioning method and a one-to-one correspondence method to detect and track the hands, respectively. The 3D positions of the head and hands were converted to the real world coordinates and the pointing position was estimated on the eye-hand ray intersecting with the large screen. Experimental results show that we improve the head detection rate applying our head segmentation method. Also, we calculate the pointing direction accuracy and the proposed method has a good performance compared with conventional methods even in dark environments.

Character Visualization in Miniature Environments with an Optical See-through Head-Mounted Display

In this paper, we present a visualization method of virtual characters to provide augmented reality (AR) experiences for a user wearing an optical see-through head-mounted display (HMD). First of all, we execute plane detection to find position of a users real desk. Second, we perform position update of virtual characters to connect real-time location information for reflecting the height of miniature objects on the desk. Finally, we visualize virtual characters that is involved in environmental properties with the optical based see-through HMD. Our method can be applied to AR contents with respect to contexts of environmental information surrounding the user such as miniature elements.