Natural scene video identification method
A technology of video recognition and natural scenes, which is applied in the field of computer vision, can solve the problems of static or dynamic, bad influence on recognition effect, and reduce the effectiveness of extracted features, so as to improve the effect of recognition
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0038] Embodiment 1: Specific steps of a natural scene video recognition method of the present invention
[0039] Such as figure 1 , 2 As shown in 3, the specific steps of a natural scene video recognition method of the present invention are as follows:
[0040] 1) Generate candidate feature point trajectories through feature point tracking, and then use the trajectory clipping method based on trajectory dissimilarity measure and ROI detection to remove trajectories caused by feature point mismatch or background changes, and finally target the reliable trajectory after clipping Calculate and extract a series of trajectory descriptors that are invariant to scale, translation, and rotation;
[0041] Among them, the steps of the trajectory pruning method as a measure of trajectory dissimilarity are as follows:
[0042] A1: Suppose there are N trajectories starting from frame f: T={t i }, i=1,...,N, for each track, define a track segment with a time window of 5 frames And adjacent frame...
Embodiment 2
[0059] Embodiment 2: Recognition effect experiment of a natural scene video recognition method of the present invention
[0060] 1. Experimental data set: including UCF sports data set and YouTube data set;
[0061] 2. Experimental environment: Matlab 2008a platform;
[0062] 3. Experimental toolbox: Kanade-Lucas-Tomasi feature tracker, VLFeat open source library and Dollar behavior recognition toolbox;
[0063] 4. Experimental method: In each experiment, first select a group of motion video sequences performed by the same actor from the sample set as test data, and use the rest of the sequences as training data. Repeat this process so that each set of motion sequences in the data set has Once used as the test data, specifically, for the YouTube data set, it is divided into 25 subsets, of which 24 subsets are used for training, and the remaining 1 subset is used for testing; for the UCF sports data set, 1 of them is video The fragments are used for testing and the rest are used for t...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 