• Journal of Korea Multimedia Society
• /
• v.20 no.6
• /
• pp.893-910
• /
• 2017

In this paper, we propose a simple online multiple object (human) tracking method, LKDeep (Lucas-Kanade feature and Detection based Simple Online Multiple Object Tracker), which can run in fast online enough on CPU core only with acceptable tracking performance for embedded surveillance purpose. The proposed LKDeep is a pragmatic hybrid approach which tracks multiple objects (humans) mainly based on LK features but is compensated by detection on periodic times or on necessity times. Compared to other state-of-the-art multiple object tracking methods based on 'Tracking-By-Detection (TBD)' approach, the proposed LKDeep is faster since it does not have to detect object on every frame and it utilizes simple association rule, but it shows a good object tracking performance. Through experiments in comparison with other multiple object tracking (MOT) methods using the public DPM detector among online state-of-the-art MOT methods reported in MOT challenge [1], it is shown that the proposed simple online MOT method, LKDeep runs faster but with good tracking performance for surveillance purpose. It is further observed through single object tracking (SOT) visual tracker benchmark experiment [2] that LKDeep with an optimized deep learning detector can run in online fast with comparable tracking performance to other state-of-the-art SOT methods.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.4
• /
• pp.1066-1079
• /
• 2023

Multi-target tracking based on the detector is a very hot and important research topic in target tracking. It mainly includes two closely related processes, namely target detection and target tracking. Where target detection is responsible for detecting the exact position of the target, while target tracking monitors the temporal and spatial changes of the target. With the improvement of the detector, the tracking performance has reached a new level. The problem that always exists in the research of target tracking is the problem that occurs again after the target is occluded during tracking. Based on this question, this paper proposes a DeepSORT model based on SIFT features to improve ship tracking. Unlike previous feature extraction networks, SIFT algorithm does not require the characteristics of pre-training learning objectives and can be used in ship tracking quickly. At the same time, we improve and test the matching method of our model to find a balance between tracking accuracy and tracking speed. Experiments show that the model can get more ideal results.

• Proceedings of the KIEE Conference
• /
• 2005.10b
• /
• pp.6-8
• /
• 2005

Study that uses geometrical information in computer vision is lively. Problem that should be preceded is matching problem before studying. Feature point should be extracted for well matching. There are a lot of methods that extract feature point from former days are studied. Because problem does not exist algorithm that is applied for all images, it is a hot water. Specially, it is not easy to find feature point in endoscope image. The big problem can not decide easily a point that is predicted feature point as can know even if see endoscope image as eyes. Also, accuracy of matching problem can be decided after number of feature points is enough and also distributed on whole image. In this paper studied algorithm that can apply to endoscope image. SIFT method displayed excellent performance when compared with alternative way (Affine invariant point detector etc.) in general image but SIFT parameter that used in general image can't apply to endoscope image. The gual of this paper is abstraction of feature point on endoscope image that controlled by contrast threshold and curvature threshold among the parameters for applying SIFT method on endoscope image. Studied about method that feature points can have good distribution and control number of feature point than traditional alternative way by controlling the parameters on experiment result.

• Phonetics and Speech Sciences
• /
• v.1 no.4
• /
• pp.111-117
• /
• 2009

In this paper, a new speech endpoint detector in noisy environment is proposed. According to the previous research, the energy feature in the speech region is easily distinguished from that in the speech absent region. In conventional method, the endpoint can be found by applying the edge detection filter that finds the abrupt changing point in feature domain. However, since the frame energy feature is unstable in noisy environment, the accurate edge detection is not possible. Therefore, in this paper, the novel feature extraction method based on spectrum envelop pattern is proposed. Then, the edge detection filter is applied to the proposed feature for detection of the endpoint. The experiments are performed in the car noise environment and a substantial improvement was obtained over the conventional method.

• Journal of Korean Society for Geospatial Information Science
• /
• v.16 no.2
• /
• pp.49-56
• /
• 2008

Image registration is one of the critical techniques of image mosaic which has many applications such as generating panoramas, video monitoring, image rendering and reconstruction, etc. The fundamental tasks of image registration are point features extraction and tracking which take much computation time. KLT(Kanade-Lucas-Tomasi) feature tracker has proposed for extracting and tracking features through image sequences. The aim of this study is to demonstrate the usage of effective and robust KLT feature detector and tracker for an image registration using the sequence image frames captured by UAV video camera. In result, by using iterative implementation of the KLT tracker, the features extracted from the first frame of image sequences could be successfully tracked through all frames. The process of feature tracking in the various frames with rotation, translation and small scaling could be improved by a careful choice of the process condition and KLT pyramid implementation.

• Journal of Korea Game Society
• /
• v.11 no.2
• /
• pp.161-166
• /
• 2011

In this paper, we propose a novel background segmentation and feature point extraction method of a human motion for the augmented reality game. First, our method transforms input image from RGB color space to HSV color space, then segments a skin colored area using double threshold of H, S value. And it also segments a moving area using the time difference images and then removes the noise of the area using the Hessian affine region detector. The skin colored area with the moving area is segmented as a human motion. Next, the feature points for the human motion are extracted by calculating the center point for each block in the previously obtained image. The experiments on various input images show that our method is capable of correct background segmentation and feature points extraction 12 frames per second.

• Journal of IKEEE
• /
• v.23 no.1
• /
• pp.207-213
• /
• 2019

In this paper, we propose a low-complexity feature detection algorithm for object tracking and present hardware architecture design and implementation results for real-time processing. The existing Shi-Tomasi algorithm shows good performance in object tracking applications, but has a high computational complexity. Therefore, we propose an efficient feature detection algorithm, which can reduce the operational complexity with the similar performance to Shi-Tomasi algorithm, and present its real-time implementation results. The proposed feature detector was implemented with 1,307 logic slices, 5 DSP 48s and 86.91Kbits memory with FPGA. In addition, it can support the real-time processing of 54fps at an operating frequency of 114MHz for $1920{\times}1080FHD$ images.

• The Journal of the Acoustical Society of Korea
• /
• v.28 no.4
• /
• pp.401-407
• /
• 2009

In this paper, we propose a novel method for the cross-correlation based double-talk detection (DTD), which employing the Gaussian Mixture Model (GMM) in the frequency domain. The proposed algorithm transforms the cross correlation coefficient used in the time domain into 16 channels in the frequency domain using the discrete fourier transform (DFT). The channels are then selected into seven feature vectors for GMM and we identify three different regions such as far-end, double-talk and near-end speech using the likelihood comparison based on those feature vectors. The presented DTD algorithm detects efficiently the double-talk regions without Voice Activity Detector which has been used in conventional cross correlation based double-talk detection. The performance of the proposed algorithm is evaluated under various conditions and yields better results compared with the conventional schemes. especially, show the robustness against detection errors resulting from the background noises or echo path change which one of the key issues in practical DTD.

• Proceedings of the SAREK Conference
• /
• 2008.06a
• /
• pp.282-287
• /
• 2008

Identification of steady-state is the first step in developing a fault detection and diagnosis (FDD) system. In a complete FDD system, the steady-state detector will be included as a module in a self-learning algorithm which enables the working system's reference model to "tune" itself to its particular installation. In this study, a steady-state detector of a residential air conditioner based on moving windows was designed. Seven representing measurements were selected as key features for steady-state detection. The optimized moving window size and the feature thresholds was suggested through startup transient test and no-fault steady-state test. Performance of the steady-state detector was verified during indoor load change test. From the research, the general methodology to design a moving window steady-state detector was provided for vapor compression applications.

• Proceedings of the IEEK Conference
• /
• 2008.06a
• /
• pp.1055-1056
• /
• 2008

In this paper, we survey various Robust Object Recognition Algorithms. One of the core technologies for local feature detector is Scale Invariant Feature Transform. And we compared several algorithms with SIFT based on IPP technology. As a result, the conversion of source codes using IPP is sped up. And this will be more improved recognition speed using SIMD Instructions.