Wangsiripitak, Somkiat
Loading...
Preferred name
Wangsiripitak, Somkiat
Main Affiliation
Email
somkiat.wa@kmitl.ac.th
21 results
Now showing 1 - 10 of 21
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Feature-based motion detection and tracking on approximate 3D ground plane(2017-01-04); Pluempitiwiriyawej, CharnchaiThe success of movement detection based on the distance moved in a 2D image sequence depends highly on the angle between a camera's optical axis and the normal vector of the ground plane on which the moving object is traveling. When the same 3D displacement occurs at various positions in the scene, the higher the angle is, the greater the distance observed in the image at a position close to camera differs from (strictly speaking, is larger than) those happening at the far end. As a consequence, a detection failure and/or false alarm may occur if no such 3D geometry is utilized. This paper estimates a 3D ground plane, which is then used to measure the approximate 3D displacement of features being detected and tracked. The 3D distances are therefore available and utilized in deciding whether they are of moving objects or just blinking features caused by illumination changes. FAST points are used to enhance a real-time system. Experimental results show superior performance in tracking: a longer trace of continuous tracking, a higher number of detected moving features, earlier detection, better recall rate, no misses, and no false alarms. A SURF descriptor and FLANN matcher were utilized here, however the robustness was not much enhanced when compared to the expense of finding the best match. - Some of the metrics are blocked by yourconsent settings
Item type:Publication, In-building navigation system using a camera(2014-01-01)A top-view map of building floor is still used nowadays by a visitor for in-building navigation. Its practicality heavily relies on the user's skill in matching a current position on the map. When the map is installed at a fixed position, not held in hand, the position of the map itself can be marked in the map, allowing the user to know the current position. Searching for the map itself, however, is still a necessary task. An orientation of the user is another issue of this traditional approach. It can be solved by having one more map point matched with the corresponding physical position on the floor. People with a good sense of direction have no difficulties with this conventional in-building navigation, but neither do the others. This paper uses the PTAM (parallel tracking and mapping - one type of visual simultaneous localization and mapping, SLAM) as a base system in which a camera is the only sensor. Two map points and their corresponding real position are used for registration of the camera pose acquired from PTAM with the top-view map coordinates. The top-view map of current floor is the only thing obtained in advance. Experimental results show that the proposed monocular navigation system can be used to navigate in the building in real-time, which helps the user correctly decide the route to the destination, similar to a GPS system used for the vehicle navigation. © 2013 IEEE. - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Visual programming using flowchart(2006-12-01) - Some of the metrics are blocked by yourconsent settings
Item type:Publication, On-line lao handwritten recognition with proportional invariant feature(2005-12-01) - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Tracking-based human entry/exit detection on various video resolutions (A study on parameter effects)(2015-01-01); A real-time tracking-based change detection using FAST features and a background feature model are proposed as a base system for detection of human entrance and exit. A speedy FAST feature extraction and tracking has to trade-off its accuracy, which sometimes causes a failure in human entrance/exit detection. Many video sizes are therefore tested in the system to examine the trade-off effects on the accuracy of feature extraction, tracking, and entry/exit detection. Tracking parameters are also investigated to determine the optimal values for each video resolution, such that stable tracking and detection are achieved. Experimental results show that the higher the video resolution is the more the error is likely to happen. Instability of feature extraction and position which increases in higher resolution is proved to be the main reason of failure. Increasing the number of previous images used in the update of background feature model, proportional to the resolution of video, takes into account the feature uncertainty. As the result, the proposed method is robust to changes in video resolution and runs at 30 fps without a miss of human entrance/exit detection and false alarm. - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Vision-Based Human Movement Matching for Muay Thai Training Support System(2024-01-01); - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Real-time monocular human height estimation using bimodal background subtraction(2017-12-19); Saelao, WongsatornA human height is one property used in conjunction with others for person identification. The method of human height estimation, using only one camera and some simple settings on the floor, is proposed to automatically determine the vertical distance of human head from the ground in real-time. Bimodal background subtraction technique helps locate the position of head top and lower foot bottom in each image frame, and the upright distance from the lower foot to the position closest to the head top is defined as the candidate of human height in the corresponding frame. A distribution of all heights estimated so far is finally used to determine the height of human in that frame. The proposed method establishes the range of possible heights from all frames in real-time - this information potentially helps increase the performance of automatic person identification in surveillance system. - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Real-Time Vision Based Human Height Measurement Using Sliding Window on Selected Candidates(2018-07-02) - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Hybrid Training of Speaker and Sentence Models for One-Shot Lip Password(2020-01-01); Lip movement can be used as an alternative approach for biometric authentication. We describe a novel method for lip password authentication, using end-to-end 3D convolution and bidirectional long-short term memory. By employing triplet loss to train deep neural networks and learn lip motions, representation of each class is more compact and isolated: less classification error is achieved on one-shot learning of new users with our baseline approach. We further introduce a hybrid model, which combines features from two different models; a lip reading model that learns what phrases uttered by the speaker and a speaker authentication model that learns the identity of the speaker. On a publicly available dataset, AV Digits, we show that our hybrid model achieved an 9.0% equal error rate, improving on 15.5% with the baseline approach. - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Improvement of Text-Independent Speaker Verification Using Gender-like Feature(2021-01-21); Text-independent speaker verification is a task of verifying a speaker identity from a characteristic of voice. We proposed the combined deep Convolutional Neural Network (CNN) consisting of (i) the first CNN trained to achieve gender classification which is then used to create a gender-like embedding and (ii) the last CNN trained with one additional input, the gender-like feature (embedding) from the first, to classify each speaker. The classification layer of the last CNN is removed to allow the remaining combined deep CNN for one-shot learning and verification of unobserved speaker. Our proposed CNN could obtain better results compared to VGGVox (ResNet-50) by 0.40% of Equal Error Rate (EER) on average. Additionally, we investigated results based on the scenario that the gender is known; the evaluation was performed only on utterance pairs that comply with the scenario. The EER rate of such case that only gender of claimed identity is known is 0.52% lower than that of VGGVox (ResNet-50) on average of two genders. In a more specific situation that the gender of person making a claim is also known, two dedicated networks were retrained for female and male, in addition to our first network which was trained for both. It is interesting that, when compared to the first network, the female network achieved less EER on female-female verification, while the network dedicated for male performed worse. Nevertheless, our two dedicated networks outperformed VGGVox (ResNet-50) by 0.88% of EER.