Methods for context-aware object tracking
The primary purpose of the present invention is to enable devices/machines/systems to perform an optimized video analytics on images and videos. The present invention focuses on detecting, tracking, and classifying objects in a scene. Here, the tracking is performed by comparing objects across at least two frames and then associating the objects based on a cost matrix. Some examples of the objects include, but are not limited to, persons, animals, vehicles, or any other articles or items.
1. A method for assigning objects in one frame to candidate objects in another frame, said method comprising the steps of:
assigning a weight of a visual cost function higher than a weight of a distance and/or expected location function for at least an object in one frame and a candidate object in a different frame based on a density of objects in any one of the frames;
computing a cost measure between the object and candidate object based on weights assigned to the at least two cost measures; and
execute object assignment based on the computed cost value.
2. The method of claim 1 , wherein the cost measure is at least one of a visual similarity or dissimilarity between the object and the candidate object is determined based on at least one of a color based similarity, gradient based similarity and texture based similarity.
3. The method of claim 1 , wherein the cost measure is a physical distance between the object and the candidate object and is determined based on number of pixels between the object and the candidate object, and wherein perspective geometry information is available.
4. The method of claim 1 , wherein the cost measure is an expected location of the object in the second frame and is determined based on at least one of location of the object in the first frame, speed of the object and direction of motion of the object.
5. The method of claim 1 , further comprising the step of applying a Hungarian method to select the one or more pairs of objects.
6. The method of claim 1 , further comprising the step of optimizing a size of a two-dimensional array (cost matrix) of cost functions between a pair of objects.
7. The method of claim 1 , further comprising classifying any one of the objects in one or more categories.
8. The method of claim 7 , wherein the classifying is based on at least one feature of the object, and wherein the feature is at least one of a size, aspect ratio, location, color, Histogram of Oriented Gradient (HOG), Scale-invariant feature transform (SIFT), HAAR like features and Local Binary Pattern (LBP) of the object.
9. The method of claim 2 , wherein the weight of the visual cost measure is assigned a higher weight compared to at least one other individual cost measure in the context of crowded objects in at least one frame.
10. The method of claim 3 , wherein the weight of the distance cost measure is assigned a higher weight compared to at least one other individual cost measure in the context of sparse objects in at least one frame.