Method, electronic device and non-transitory computer readable storage medium for image annotation

An image annotation method implemented by an electronic device including a processor, comprising:

• acquiring a sequence of image frames comprising a plurality of image frames from video data by the processor;
• performing an object detecting and tracking procedure on the sequence of image frames by the processor, so as to identify and track one or more target objects from the image frames;
• selecting a plurality of candidate key image frames from the image frames according to a first selection condition by the processor, wherein the first selection condition comprises when a target object in the one or more target objects starts to appear or disappears in an image frame of the image frames, selecting the image frame as one of the candidate key image frames;
• determining a plurality of first similarity indexes of the candidate key image frames by the processor, wherein each of the first similarity indexes is determined by the processor through a similarity calculation according to a first covariance value of a corresponding one of the candidate key image frames and a plurality of first variation values statistically calculated in different directions of the corresponding candidate key image frame;
• determining a plurality of second similarity indexes of a plurality of adjacent image frames by the processor, wherein each of the adjacent image frames is adjacent to at least one of the candidate key image frames, and each of the second similarity indexes is determined by the processor through the similarity calculation according to a second covariance value of a corresponding one of the adjacent image frames and a plurality of second variation values statistically calculated in different directions of the corresponding adjacent image frame;
• selecting the candidate key image frames as well as the adjacent image frames that meet a second selection condition as a plurality of key image frames, wherein the second selection condition comprises when a difference between a corresponding second similarity index of an adjacent image frame of the adjacent image frames and a corresponding first similarity index of a candidate key image frame adjacent to the adjacent image frame exceeds a similarity threshold, selecting the adjacent image frame as one of the key image frames;
• and
• presenting the key image frames on a graphical user interface and displaying annotation information for the one or more target objects through the graphical user interface by the processor.