scispace - formally typeset
Search or ask a question
Author

Liu Ning

Bio: Liu Ning is an academic researcher from Sun Yat-sen University. The author has contributed to research in topics: Video tracking & Feature (computer vision). The author has an hindex of 3, co-authored 4 publications receiving 15 citations.

Papers
More filters
Patent
07 Dec 2016
TL;DR: In this article, a weak structure perception visual target tracking method capable of fusing with context detection is proposed, in which a target and two surrounding component sets are combined with a movement model to generate a potential target center, then, the potential target centre is clustered to reject noise to obtain an accurate target position, and a target size is updated.
Abstract: The invention discloses a weak structure perception visual target tracking method capable of fusing with context detection. During initialization, a weak structure relationship between a target and each component of surrounding environment is perceived to establish a model. Model maintenance corresponds to the target and two surrounding component sets, and a feature point and a feature descriptor are used for expressing the appearances of the components. In a tracking process, the component sets are combined with a movement model to generate a potential target center, then, the potential target center is clustered to reject noise to obtain an accurate target position, and a target size is updated. Under a weak structure tracking frame, a bottom-up way and a top-down way are introduced to carry out target context detection in order to enhance the prediction of a component position. Bottom-up detection provides consistent tracking information for each component through the estimation of the local movement of a pixel level. Top-down detection constructs a superpixel nuclear model to learn a difference between the target and a background on a level of individual, and guidance information is provided for target positioning and model update.

6 citations

Patent
31 Aug 2018
TL;DR: In this paper, a target tracking method fusing convolutional network features and a discriminant correlation filter is proposed, which is trained by learning rich stream information in continuous frames, thereby improving feature representation and tracking precision.
Abstract: The invention discloses a target tracking method fusing convolutional network features and a discriminant correlation filter. An end-to-end lightweight network system structure is established; and theconvolutional features are trained by learning rich stream information in continuous frames, thereby improving feature representation and tracking precision. A correlation filter tracking component is constructed as a special layer in a network to track a single image block; in the tracking process, a target block and multiple background blocks are tracked at the same time; by perceiving a structural relationship between a target and surrounding background blocks, a model is built for a part with a high discrimination degree for the target and a surrounding environment; a target tracking effect is measured through a relationship between a peak sidelobe ratio and a peak value of a confidence map; and under the condition of high tracking difficulty such as large-area shielding, target shapeextreme deformation, illumination drastic change, locating is performed by automatically utilizing the discriminated background part.

6 citations

Proceedings ArticleDOI
01 Dec 2016
TL;DR: The proposed method is evaluated with extensive experiments both quantitatively and qualitatively, demonstrating that it is robust to various challenging factors in object tracking and outperforms many state-of-the-art methods.
Abstract: Visual object tracking is a fundamental task in many high-level computer vision applications. Most existing algorithms have to build complex models with expensive computation to achieve accurate object tracking, which brings significant difficulty in real-time tracking. In order to address this problem, motivated by recent success of high-speed correlation filter (CF) models, a novel real-time object tracking method is proposed in this paper based on multi-cue adaptive correlation filters (MCF). Different from conventional CF-based trackers that employ a fixed feature, our method explores multi-cue adaptation that automatically discovers the effective cues to handle specific scenarios. Specifically, our framework is built with multiple CF-based trackers using different cues, such as intensity, color, texture and edge. In a new frame, the trackers run their own tasks simultaneously and compete to locate the target. A unified measure is introduced to evaluate the tracking performance of different trackers and find out the best prediction, which will be used as the final result of all the trackers. The proposed method is evaluated with extensive experiments both quantitatively and qualitatively, demonstrating that it is robust to various challenging factors in object tracking and outperforms many state-of-the-art methods.

4 citations

Journal ArticleDOI
TL;DR: A novel method for modeling and locating the object by being aware of the weak structures of discriminative parts of both the object and its surroundings and builds a superpixel kernel model to roughly distinguish the object from its surroundings, which provides guided information for location inference and model update.
Abstract: It poses great challenges to model-free trackers that the object undergoes large appearance variations due to motion, shape deformation, occlusion and surrounding environments. In this paper, we investigate a novel method for modeling and locating the object by being aware of the weak structures of discriminative parts of both the object and its surroundings. The discriminative parts are modeled based on keypoints and feature descriptors. We separate the discriminative parts into two sets corresponding to object and background, and model their spatial structure relationship with the object. While tracking, the successfully localized parts will contribute to potential centers of the object. Aware of the weak structures, we further cluster potential centers to locate the object. The object scale is also updated adaptively. To increase the accuracy of this weak-structure-aware location inference, we fully explore context in both bottom-up and top-down procedures. In the bottom-up stage, we explore the local motion estimation of low-level pixels. The bottom-up information produces consistent tracking of discriminative parts. In the top-down stage, we build a superpixel kernel model to roughly distinguish the object from its surroundings, which provides guided information for location inference and model update. The effectiveness of the proposed method is verified by evaluation on a popular benchmark and comparison with recent tracking methods.

Cited by
More filters
Patent
31 May 2017
TL;DR: In this article, a moving target tracking method based on an optical flow method and key point features is proposed, which is high in calculation speed and highly resistant to occlusion and deformation.
Abstract: The invention discloses a moving target tracking method based on an optical flow method and key point features. The moving target tracking method includes: in initial video frames, calculating feature vectors corresponding to target area key points and background area key points and establishing a feature library on the basis of the feature vectors; using the optical flow method to remove the unstable key points between every two neighboring frames to acquire the key points succeeding in optical flow tracking; detecting and describing all key point features in the current frames and matching the key point features with the feature library to acquire multiple optimum matched key points; fusing the key points succeeding in tracking and the successfully matched key points; evaluating a central position, a size and a rotary angle of a target according to a similar triangle relation; updating the feature library on line by the aid of historical frame information. The moving target tracking method has the advantages that by the moving target tracking method, long-time stable tracking on the target can be realized, and real-time geometric status information of the target size, the rotary angle and the like can be accurately evaluated; the moving target tracking method is high in calculation speed and highly resistant to occlusion and deformation.

9 citations

Patent
13 Jun 2017
TL;DR: In this paper, a context sensing-based automatic auxiliary driving system is described, where an information node audiomonitor is used to monitor updating of distributed information nodes and stored context information; a context retriever is in charge of indexing and retrieving the stored contexts; and when context is retrieved to be changed, a rule engine was used for searching one or more matched targets as well as is used for ratiocinating high-level context according to the rule of a database, transmitting fusion information to a control unit and controlling an actuator and man-machine interaction.
Abstract: The invention discloses a context sensing-based automatic auxiliary driving system An information node audiomonitor is in charge of monitoring updating of distributed information nodes and stored context information; a context retriever is in charge of indexing and retrieving the stored context information; an interpreter is used for providing service for the information node audiomonitor and the context retriever; and when context is retrieved to be changed, a rule engine is used for searching one or more matched targets as well as is used for ratiocinating high-level context according to the rule of a database, transmitting fusion information to a control unit and controlling an actuator and man-machine interaction to enhance automatic auxiliary driving

6 citations

Patent
31 Aug 2018
TL;DR: In this paper, a target tracking method fusing convolutional network features and a discriminant correlation filter is proposed, which is trained by learning rich stream information in continuous frames, thereby improving feature representation and tracking precision.
Abstract: The invention discloses a target tracking method fusing convolutional network features and a discriminant correlation filter. An end-to-end lightweight network system structure is established; and theconvolutional features are trained by learning rich stream information in continuous frames, thereby improving feature representation and tracking precision. A correlation filter tracking component is constructed as a special layer in a network to track a single image block; in the tracking process, a target block and multiple background blocks are tracked at the same time; by perceiving a structural relationship between a target and surrounding background blocks, a model is built for a part with a high discrimination degree for the target and a surrounding environment; a target tracking effect is measured through a relationship between a peak sidelobe ratio and a peak value of a confidence map; and under the condition of high tracking difficulty such as large-area shielding, target shapeextreme deformation, illumination drastic change, locating is performed by automatically utilizing the discriminated background part.

6 citations

Patent
14 Jun 2019
TL;DR: In this article, an image processing method and device, a storage medium, equipment and a system belonging to the technical field of machine learning is described, which comprises the steps of obtaining a video image stream of a to-be-detected body part; sequentially carrying out focus detection on each frame of image in the video image streams; classifying the current frame of images according to a first focus detection result of a previous frame and a second focus detection results of the current one; wherein the previous frame of an image is at least one frame of the image located in front
Abstract: The invention discloses an image processing method and device, a storage medium, equipment and a system, and belongs to the technical field of machine learning. The method comprises the steps of obtaining a video image stream of a to-be-detected body part; sequentially carrying out focus detection on each frame of image in the video image stream; for the current frame of image, classifying the current frame of image according to a first focus detection result of a previous frame of image and a second focus detection result of the current frame of image; wherein the previous frame of image is at least one frame of image located in front of the current frame of image in time sequence. The invention discloses an image processing method. The prediction result of the previous frame of image isconsidered in the prediction of the current frame of image, the advantages of high efficiency and no accumulative error of a single frame of image detection method are integrated, the accuracy of image classification is remarkably improved by fusing the related information of other frames of images, and the continuity of the prediction result is ensured.

4 citations

Journal ArticleDOI
TL;DR: In this article, a remote sensing video satellite multiple object detection and tracking method based on road masking, Gaussian mixture model (GMM), and data association is proposed, which extracts the road network from the remote sensing videos based on deep learning.
Abstract: A remote sensing video satellite multiple object detection and tracking method based on road masking, Gaussian mixture model (GMM), and data association is proposed. This method first extracts the road network from the remote sensing video based on deep learning. In the detection stage, the background subtraction algorithm is used based on the GMM to obtain the detection results of the moving targets on the road. In the tracking stage, the data association of the same target detection result in adjacent frames is realized based on the neighborhood search algorithm, so as to obtain the continuous tracking trajectory of each target. The experiments about multiobject detection and tracking are conducted on data measure by real remote sensing satellites, and the results verified the feasibility of the proposed method.

4 citations