Abstract: There is provided a method, device, and system for tracking an object in a sequence of video frames by using a tracking filter which is operable in two modes. In the first mode it estimates a state vector describing a position and velocity of the object in the video frames. In the second mode it estimates a state vector describing position, velocity, and acceleration of the object in the video frames. The method includes switching from the first mode to the second mode. The switching is performed in response to a velocity difference between a velocity of the object estimated by the tracking filter and a velocity of the object indicated by motion vectors from a video encoder and being larger than a threshold.

Abstract: A method for feature extraction of detected objects, comprising the steps of: receiving a plurality of images, each depicting an object detected by the object detecting application; concatenating the plurality of images into a composite image according to a grid pattern; feeding the composite image through a convolutional neural network (CNN) trained for feature extraction, wherein each convolutional layer of the CNN is configured to, while convolving input data to the convolutional layer using a convolutional filter: determine a currently convolved image of the plurality of images by determining a centre coordinate of a subset of the input data currently covered by the convolutional filter, and mapping the centre coordinate to the grid pattern; and selectively nullifying all weights of the convolutional filter that cover input data derived from any of the plurality of images not being the currently convolved image.

Abstract: A method, system and software for controlling a list of inactive tracks in an object tracking system. The techniques described includes obtaining the false positive rate, FPR, of a re-identification model using a decision threshold, and obtaining an acceptable FPR of the tracking system. The re-identification model is used to match new detections with inactive tracks. The techniques include determining one or more terminal conditions for deleting inactive tracks based on the acceptable and model FPRs. If a first inactive track meets a terminal condition, it is deleted from the list.

Abstract: A computer-implemented method of tracking objects in a video sequence of a scene comprises determining a location of a sink in the scene where objects exit the scene and a location of a source where objects enter the scene; tracking a first object moving in the scene using a re-identification algorithm, wherein the first object is associated with a re-identification threshold of the re-identification algorithm; detecting that the first object has exited the scene at the sink; and responsive to detecting that the first object has exited the scene at the sink, adjusting the re-identification threshold associated with the first object such that a probability that the re-identification algorithm re-identifies a second object, entering the scene at the source after the first object has exited the scene at the sink, as the first object is reduced.

Abstract: An apparatus and a method for tracking a temporarily occluded object in image frames of objects and associating detected instances into tracks. An active first track of a first object and an active second track of a second object are identified, which objects are moving together. Upon determination that the first track, but not the second track, has become inactive connecting it to the second track. Thereafter, a detected object instance is associated with an inactive track when a similarity score between the detected instance and the inactive track fulfills a similarity requirement. The requirement is less strict when the instance is spatially proximate to the second track and when the detected instance is compared to the inactive first track connected to the second track as compared to when the detected instance is compared to an unconnected inactive track.

Abstract: A method of encoding images in a video, comprises: acquiring an original image from an image sensor of a video camera; encoding the original image using a generative image model, thereby obtaining a first encoded image; decoding the image to obtain a first decoded image; identify a region of interest (ROI) of the original image; for each ROI: perform an encoding quality check by comparing several reference points in the ROI of the original image against corresponding reference points in the ROI of the first decoded image, thereby obtaining a difference; if the difference is greater than a threshold, encoding the ROI using a non-generative image model, thereby obtaining a non-generative encoded image area; providing final encoded image data comprising a) the non-generative encoded image areas for the ROI having a difference greater than the threshold and b) the first encoded image for a remaining part of the original image.

Abstract: A method and a device for setting a value of an object property in a sequence of metadata frames corresponding to a sequence of video frames is provided. An object is detected in a plurality of video frames. For a temporally first video frame a first value of an object property is determined and the object property is set to have the first value in a metadata frame corresponding to the first video frame. For each subsequent video frame, a subsequent value of the object property is determined and a deviation is calculated with respect to a value of the object property that was last set in a metadata frame. If the deviation exceeds a non-zero threshold, the object property is set to have the subsequent value in a metadata frame corresponding to the subsequent video frame. Otherwise no value of the object property is set in that metadata frame.

Abstract: A plurality of feature vectors for moveable objects are received in a sequence of image frames captured during a first period. The feature vectors are received from a machine learning module trained to extract similar feature vectors in different image frames. An initial value is assigned to a first feature vector , or to a cluster of feature vectors identified in a second sequence of image frames preceding the first image frame. The indicator indicates whether vector is alive. For each subsequent image frame , the indicator is updated by: updating the value based on a similarity between the feature vector and the first feature vector or the cluster of feature vectors. If the value of the indicator indicates that the vector is alive, determining that there is a moveable object that is located in the captured scene at least the predetermined portion of the given period of time.

Abstract: The present disclosure generally relates to a method for weighting of features in a feature vector of an object detected in a video stream capturing a scene, comprising: determining a feature vector comprising a set of features for a detected object in the video stream; acquiring a reference feature vector of a reference model of the scene; and assigning a weight to at least one feature of the determined feature vector, wherein the weight for a feature of the determined feature vector depends on a deviation measure indicative of a degree of deviation of the feature from a corresponding feature of the acquired reference feature vector of the reference model.

Abstract: A method for tracking objects in a scene, comprises detecting object candidates in an image frame, and calculating, for each object candidate, an association measure which is indicative of a likelihood that the object candidate is associated with a current object track. Additionally, a view of the overview camera is correlated with a heatmap of the scene, the heatmap providing data indicative of areas in the scene having an elevated degree of occurrence of historical verified object tracks, and adjusting the association measure or an association threshold for object candidates which according to the heatmap are located in areas in the scene having an elevated degree of occurrence of historical verified object tracks so as to increase their probability of being associated with a current object track. Each object candidate is then associated with a current object track if the association measure is above an association threshold.

Abstract: A method for determining similarity in appearance of objects in image frames of a video sequence, comprises: processing first raw image data applying first image processing settings; detecting a first object; extracting first feature vectors; processing second raw image data applying second image processing settings; detecting a second object; extracting second feature vectors; determining the similarity of the first and second objects by comparing the first feature vectors to the second feature vectors; if the first feature vectors and the second feature vectors differ by more than a first threshold, and the first image processing settings differ from the second image processing settings by more than a second threshold: re-processing the first raw image data corresponding applying the second image processing settings; extracting, from the first image area, updated first feature vectors; comparing the updated first feature vectors to the one or more second feature vectors.

Abstract: A device, a non-transitory computer-readable storage medium, and a method of signing a metadata frame corresponding to an image frame of a sequence of image frames are disclosed. The metadata frame comprises metadata of one or more detected objects in the image frame, and the metadata of each detected object comprises coordinates defining a location in the image frame of the detected object. A digital signature is generated based on at least a subset of the metadata in the metadata frame, and additional metadata are added to the metadata frame. The additional metadata comprise the digital signature and predefined coordinates which define that the additional metadata comprise the digital signature. Furthermore, a device, a non-transitory computer-readable storage medium, and method of authenticating a digitally signed metadata frame corresponding to an image frame of a sequence of image frames are disclosed.

Abstract: A computer-implemented method of tracking objects in a video sequence of a scene comprises determining a location of a sink in the scene where objects exit the scene and a location of a source where objects enter the scene; tracking a first object moving in the scene using a re-identification algorithm, wherein the first object is associated with a re-identification threshold of the re-identification algorithm; detecting that the first object has exited the scene at the sink; and responsive to detecting that the first object has exited the scene at the sink, adjusting the re-identification threshold associated with the first object such that a probability that the re-identification algorithm re-identifies a second object, entering the scene at the source after the first object has exited the scene at the sink, as the first object is reduced.

Abstract: A method for prioritizing feature extraction for object re-identification in an object tracking application. Region of interests (ROI) for object feature extraction is determined based on motion areas in the image frame. Each object detected in an image frame and which is at least partly overlapping with a ROI is associated with the ROI. A list of candidate objects for feature extraction is determined by, for each ROI associated with two or more objects: adding each object of the two or more objects that is not overlapping with any of the other objects among the two or more objects with more than a threshold amount. From the list of candidate objects, at least one object is selected, and image data of the image frame depicting the selected object is used for determining a feature vector for the selected object.

Abstract: A method of encoding images in a video, comprises: acquiring an original image from an image sensor of a video camera; encoding the original image using a generative image model, thereby obtaining a first encoded image; decoding the image to obtain a first decoded image; identify a region of interest (ROI) of the original image; for each ROI: perform an encoding quality check by comparing several reference points in the ROI of the original image against corresponding reference points in the ROI of the first decoded image, thereby obtaining a difference; if the difference is greater than a threshold, encoding the ROI using a non-generative image model, thereby obtaining a non-generative encoded image area; providing final encoded image data comprising a) the non-generative encoded image areas for the ROI having a difference greater than the threshold and b) the first encoded image for a remaining part of the original image.

Abstract: A method for selecting a crop score threshold for enhancing tracking of objects in a scene captured in a video sequence is disclosed. A respective track is obtained for two different objects, each track comprising crops of object instances of the objects in in a video sequence, each crop having a crop score and a feature vector. Each track is split into respective more tracklets thereby forming four or more tracklets. For each candidate crop score threshold a respective difference between each tracklet and each other tracklet is determined based on differences between feature vectors of crops having a crop score above the candidate crop score threshold of each tracklet, and each other tracklet. A crop score threshold is selected from the set of crop score thresholds resulting in a maximum difference between the differences between tracklets of different tracks and the differences between tracklets of the same track.

Abstract: A method detects one or more occluded areas of a scene analysed by an object tracking system. The method includes building a map of one or more occluded areas in a scene. Building the map comprises running a re-identification algorithm on a video sequence to try to resume a lost object track. If the object track is successfully resumed, the method includes determining an area of the scene where the first object track is lost and an area of the scene where the first object track is resumed. A connection between the first and the second area of the scene is added the map such that the map identifies that an object track being lost in the first area of the scene has been resumed in the second area of the scene.

Abstract: A method for feature extraction of detected objects, comprising the steps of: receiving a plurality of images, each depicting an object detected by the object detecting application; concatenating the plurality of images into a composite image according to a grid pattern; feeding the composite image through a convolutional neural network (CNN) trained for feature extraction, wherein each convolutional layer of the CNN is configured to, while convolving input data to the convolutional layer using a convolutional filter: determine a currently convolved image of the plurality of images by determining a centre coordinate of a subset of the input data currently covered by the convolutional filter, and mapping the centre coordinate to the grid pattern; and selectively nullifying all weights of the convolutional filter that cover input data derived from any of the plurality of images not being the currently convolved image.

Abstract: The present system and method generally relate to the field of camera surveillance, and in particular to object re-identification in video streams captured by a camera.

Abstract: A method of detecting a change of ratio of occlusion of a tracked object in a video sequence. For each of a plurality of image frames, a bounding box of the tracked object is determined and for each pair of successive image frames, an intersection over union (IoU) of a first bounding box in a first image frame and a second bounding box in a second image frame is calculated. Similarly, for a further pair of successive image frames, a further IoU of a first bounding box in a first image frame of the further pair of successive image frames and a second bounding box in a second image frame of the further pair of successive image frames is calculated. If the further IoU differs from the calculated IoUs by more than a threshold amount, the ratio of occlusion of the tracked object has changed.