首页|Researchers Submit Patent Application, 'Multi-Object Tracking Using Correlation Filters In Video Analytics Applications', for Approval (USPTO 20240303836)
Researchers Submit Patent Application, 'Multi-Object Tracking Using Correlation Filters In Video Analytics Applications', for Approval (USPTO 20240303836)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
News editors obtained the following quote from the background information suppli ed by the inventors: “Efficient and effective object tracking is a critical task in a visual perception pipeline, as it bridges inference results across video f rames, enabling temporal analysis of objects of interest. Tracking multiple obje cts is a key problem for many applications such as surveillance, animation, acti vity recognition, or vehicle navigation. Conventional multi-object trackers may be implemented using independent single-object trackers that run on full-frames of video and track objects by associating bounding boxes between frames. Trackin g is typically performed on a single video stream and divided into localization and data association. For localization, each single-object tracker may independe ntly estimate a location of a detected object in a frame- and for data associati on-estimated object locations from the trackers may be linked across frames to f orm complete trajectories. Discriminative Correlation Filters (DCFs) have recent ly been used for localization in object tracking. DCF-based trackers may define a search region around an object of interest, where an optimal correlation filte r is learned so that the object can be localized in the next frame as the peak l ocation of a correlation response within the search region.
NETL
NSTL
