Object Tracking with Instance Matching and Online Learning
A real-time object tracking algorithm is proposed to cope with the variables of appearance changes like translation, zooming, rotation, panning/tilting, occlusion, luminance change, and blur.
Abstract
A real-time object tracking algorithm is proposed to cope with the variables of appearance changes like translation, zooming, rotation, panning/tilting, occlusion, luminance change, and blur. The proposed tracking scheme includes three steps. First, regional filter is employed to detect the candidate regions of targets. Next, these candidate regions are scaled to an uniform size for feature extraction. Finally, using feature matching to calculate the similarity between an instance and the target, and then store this instance if recognized as the target. We can see that the instance database would contain object's difference appearances as the tracking time going on. In other words, recognition capability will increase while the database become enlarging. To keep high computation performance, an algorithm with database reduction is proposed to limit the size of database. From our experiments, the proposed tracking system can achieve 30 FPS with resolution 1280x720 on an Intel I5 CPU 2.6GHz.
Ongoing Progress
High-Dimension Feature Graph for Multi-Source Integration
Contrary to calculating feature extraction from each individual source, here we propose the high-dimension(HD) feature graph to integrate the extraction results from multi-source. First, each individual source can build one feature graph by modeling each feature descriptor as one vertex and connecting the vertices with homogeneity. Next, multiple feature graphs are integrated to build a HD feature graph. The proposed HD feature graph can achieve higher recognition accuracy benefit from modeling various feature descriptors into a higher dimension feature graph, which not only preserves heterogeneity feature descriptors but also records the features’ spatial correlation.
Object Identification from A Large Database
Object identification from a large database is absolutely a difficult challenge for real-time applications. To overcome this challenge, it not only requires a fast matching algorithm but also needs an efficient database system. Here, we propose the HD graph matching algorithm which is able to calculate graph similarity faster than linear feature matching. Next, a specified graph database architecture is proposed to stored HD graph for efficient graph access and computation. With the innovation from algorithms and database structures, object identification from the large database is practical for real-time applications.
Patents
Abstract
An image tracking device and an image tracking method thereof are provided. The image tracking device includes an image capture interface, a storage means, and a processor means. The storage means has a multi-dimensional storage space for storing a plurality of first images, each dimension of the multi-dimensional storage space being corresponding to a feature-related variance of a multi-dimensional variance. The processor means is configured to execute the following operations: marking a second image in the picture frame; calculating a multi-dimensional variance between the second image and each of the first images separately; determining whether the second image contains the object according to the multi-dimensional variance calculated; and if the second image is determined as one containing the object, storing the second image as one of the first images, in a specific subspace of the multi-dimensional storage space according to the multi-dimensional variance calculated.
Presentation Slide
Demo Videos
Our algorithm running with real-time capability.
If you cannot see the video below, click the link https://www.youtube.com/embed/XkG2N0poe1M
Demo video for rigit object tracking.
If you cannot see the video below, click the link https://www.youtube.com/embed/t2KCFfs802g
Demo video for non-rigit object tracking.
If you cannot see the video below, click the link https://www.youtube.com/embed/peLYve6znoo
Demo video for face tracking.
If you cannot see the video below, click the link https://www.youtube.com/embed/zNIMCAOJr9Y