System and method for video-quality enhancement

1. A video-quality enhancement system comprising:

an input that receives image data; and

a near-field enhancement module that identifies a near-field obscuration pixel in the received image data, determines whether the identified near-field obscuration pixel is located on a foreground object, and fills the identified near-field obscuration pixel based on the determination of whether the identified near-field obscuration pixel is located on the foreground object to generate a near-field enhanced image;

wherein the near-field enhancement module determines whether the identified near-field obscuration pixel is located on the foreground object based on calculated distances from the identified near-field obscuration pixel to each foreground object in the image data.

2. The video-quality enhancement system of claim 1 , wherein the near-field enhancement module is initialized based on a plurality of image frames that are analyzed to create a background map and an obscuration map, wherein the background map and the obscuration map are used to identify the identified near-field obscuration pixel in real-time in a current frame of image data.

3. The video-quality enhancement system of claim 1 , wherein the near field enhancement module further includes:

a temporal object detector that analyzes successive frames of image data to determine whether the identified near-field obscuration pixel is located on a background or foreground object.

4. The video-quality enhancement system of claim 1 , wherein the identified near-field obscuration pixel identified by the near-field obscuration module includes a snowflake/raindrop pixel identified in the near-field of the image data.

5. The video-quality enhancement system of claim 1 , further including:

an adaptive far-field enhancement module that divides the near-field enhanced image provided by the near-field enhancement module into a plurality of tiles, calculates a texture value associated with each of the plurality of tiles, and applies local enhancement to each of the plurality of tiles based on the calculated texture to generate a far-field enhanced image.

6. The video-quality enhancement system of claim 5 , wherein the adaptive far-field enhancement module calculates a global blockiness value with the image data and applies local enhancement to each of the plurality of tiles based on the calculated global blockiness value.

7. A video-quality enhancement system comprising:

an input that receives image data; and

a near-field enhancement module that identifies a near-field obscuration pixel in the received image data, determines whether the identified near-field obscuration pixel is located on a foreground object, and fills the identified near-field obscuration pixel based on the determination of whether the identified near-field obscuration pixel is located on the foreground object to generate a near-field enhanced image;

wherein the near field enhancement module further includes a temporal object detector that analyzes successive frames of image data to determine whether the identified near-field obscuration pixel is located on a background or the foreground object;

wherein the temporal object detector determines whether the identified near-field obscuration pixel is located on the foreground object based on calculated minimum distances from the identified near-field obscuration pixel to each foreground object and variance of the calculated minimum distances over successive frames of image data.

8. The video-quality enhancement system of claim 7 , wherein the temporal object detector determines whether the identified near-field obscuration pixel is located on the foreground object based on a minimum distance from the identified near-field obscuration pixel to a nearest foreground object and a calculated variance associated with the minimum distance from the identified near-field obscuration pixel to a nearest foreground object in each successive frame.

9. A method of enhancing video-quality for improved computer vision, the method comprising:

receiving image data from a video detector;

identifying a near-field obscuration pixel within the image data;

determining whether the identified near-field obscuration pixel is located on a foreground object within the image data; and

creating a near-field enhanced image from the image data by filling the identified near-field obscuration pixel with foreground painting if the identified near-field obscuration pixel is located over a foreground object, otherwise, filling the identified near-field obscuration pixel based on a background image;

wherein determining whether the identified near-field obscuration pixel is located on the foreground object is in response to calculated distances from the identified near-field obscuration pixel to each foreground object.

10. The method of claim 9 , wherein identifying the identified near-field obscuration pixel includes subtracting a background map and an obscuration map from a current image.

11. The method of claim 9 , wherein the identified near-field obscuration pixel identified within the image data includes those pixels identified as snowflake/raindrop pixels.

12. The method of claim 9 , further including:

dividing the near field enhanced image into a plurality of tiles;

calculating texture values for each of the plurality of tiles and

creating a far-field enhanced image based, in part, on the calculated texture value for each of the plurality of tiles.

13. The method of claim 12 , further including:

calculating a global blockiness value associated with the near-field enhanced image; and

creating the far field enhanced image based, in addition, on the calculated global blockiness value.

14. A method of enhancing video-quality for improved computer vision, the method comprising:

receiving image data from a video detector;

identifying a near-field obscuration pixel within the image data;

determining whether the identified near-field obscuration pixel is located on a foreground object within the image data; and

creating a near-field enhanced image from the image data by filling the identified near-field obscuration pixel with foreground painting if the identified near-field obscuration pixel is located over the foreground object, otherwise, filling the identified near-field obscuration pixel based on a background image;

wherein determining whether the identified near-field obscuration pixel is located on the foreground object includes calculating distances from the identified near-field obscuration pixel to a center of each foreground object identified in a plurality of images;

selecting a minimum distance for each identified near-field obscuration pixel to a nearest foreground object in the plurality of images;

calculating variances associated with the minimum distance selected for each identified near-field obscuration pixel over the plurality of images; and

determining whether the identified near-field obscuration pixel is located on the foreground object based on the selected minimum distance and the calculated variance.

15. A video-quality enhancement system for snow/rain image enhancement, the system comprising:

means for receiving image data from a video detector;

means for identifying a near-field snowflake/raindrop pixel;

means for determining whether the identified near-field snowflake/raindrop pixel is located on a foreground object; and

means for creating a near-field enhanced image by filling the identified near-field snowflake/raindrop pixel with foreground inpainting if the identified near-field snowflake/raindrop pixel is located over a foreground object, otherwise, filling the identified near-field snowflake/raindrop pixel based on a background image;

wherein the means for determining whether the identified near-field snowflake/raindrop pixel is located on the foreground object determines whether the identified near-field obscuration pixel is located on the foreground object in response to calculated distances from the identified near-field obscuration pixel to each foreground object.

16. The video-quality enhancement system of claim 15 , wherein the means for identifying the near-field snowflake/raindrop pixel subtracts a stored background map and a stored snow/rain map from a current image to identify the near-field snowflake/raindrop pixel.

17. The video-quality enhancement system of claim 16 , further including:

means for dividing the near-field enhanced image into a plurality of tiles;

means for calculating texture values for each of the plurality of tiles; and

means for creating a far-field enhanced image based, in part, on the calculated texture value for each of the plurality of tiles.

18. The video-quality enhancement system of claim 17 , further including:

means for calculating a global blockiness value associated with the near-field enhanced image; and

means for modifying the far-field enhanced image based, in addition, on the calculated global blockiness value.

19. A video-quality enhancement system for snow/rain image enhancement, the system comprising:

means for receiving image data from a video detector;

means for identifying a near-field snowflake/raindrop pixel;

means for determining whether the identified near-field snowflake/raindrop pixel is located on a foreground object; and

means for creating a near-field enhanced image by filling the identified near-field snowflake/raindrop pixels with foreground inpainting if the identified near-field snowflake/raindrop pixel is located over a foreground object, otherwise, filling the identified near-field snowflake/raindrop pixel based on a background image;

wherein the means for identifying the near-field snowflake/raindrop pixel subtracts a stored background map and a stored snow/rain map from a current image to identify the identified near-field snowflake/raindrop pixel;

wherein the means for determining whether the identified near-field snowflake/raindrop pixel is located on the foreground object includes:

means for calculating distances from the identified near-field snowflake/raindrop pixel to a center of each foreground object identified in a plurality of images;

means for selecting a minimum distance for each identified near-field snowflake/raindrop pixel to a nearest foreground object in the plurality of images;

means for calculating variances associated with the minimum distance selected for each identified near-field snowflake/raindrop pixel over the plurality of images; and

means for determining whether the identified near-field snowflake/raindrop pixel is located on the foreground object based on the selected minimum distance and the calculated variance.