首页|"Device And Method For Enhancing Tracking Of Objects In A Scene Captured In A Vi deo Sequence" in Patent Application Approval Process (USPTO 20240303828)
"Device And Method For Enhancing Tracking Of Objects In A Scene Captured In A Vi deo Sequence" in Patent Application Approval Process (USPTO 20240303828)
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
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News-A patent application by the inventors Danielsson, Niclas (Lund, SE); Skans, Markus (Lund, SE); Ohrn, Anton (Lund, SE), filed on February 12, 2024, was made available online on September 12, 2024, ac cording to news reporting originating from Washington, D.C., by NewsRx correspon dents. This patent application is assigned to Axis AB (Lund, Sweden). The following quote was obtained by the news editors from the background informa tion supplied by the inventors: "In prior art, when tracking objects in a scene captured in a video sequence, tracklets are identified including crops from a su b-set of subsequent image frames of the video sequence. Each crop of a tracklet is a portion of an image frame of the video sequence and is associated with a de tection in the image frame. For example, each crop may be the portion of a corre sponding image frame that is within a bounding box around a detected object in the image frame. All crops of a tracklet have been determined to relate to the sa me object according to a given criterion. Tracklets belonging to the same object should then be connected to produce a single track relating to that same object . To determine if two different tracklets belong to the same tracked object, fea ture vectors of the crops of the respective tracklet may be used. Such feature v ectors may for example be feature vectors determined by means of a convolutional neural network. SpecificAlly, a feature vector may be determined for each of th e two tracklets, e.g. as average of the feature vectors of the crops of the resp ective tracklet, and the tracklets may then be determined as belonging to the sa me tracked object if the distance between their feature vectors is less than a t hreshold. A problem that may arise in such prior art schemes for connecting trac klets is that the determined feature vector for the tracklets may not be represe ntative for the object and hence errors may occur such as two tracklets being de termined not to relate to the same object even if they actuAlly do."