Detection of near-field occlusions in images
A method is presented for detecting occlusions on a color camera. The method includes: capturing an image of a scene using the camera; analyzing intensity of electromagnetic radiation forming the at least one image, where the intensity of the electromagnetic radiation is analyzed across the electromagnetic spectrum; detecting an occlusion on a lens of the camera based on variation of intensity of the electromagnetic radiation across the electromagnetic spectrum; and tagging the image with an indicator of the occlusion, where the tagging occurs in response to detecting an occlusion on the camera.
1. A method for detecting occlusions on a color camera, comprising:
capturing, by the camera, an image of a scene, where electromagnetic radiation captured by the camera is captured in at least two mutually distinct frequency bands;
comparing, by an image processor, intensity of electromagnetic radiation in one frequency band to intensity of electromagnetic radiation in another frequency band, where the one frequency band is distinct from the another frequency band and the electromagnetic radiation for the one frequency band is taken from same image as the electromagnetic radiation for the another frequency band; and
detecting, by the image processor, an occlusion on a lens of the camera when the intensity of the electromagnetic radiation in the one frequency band is similar to the intensity of the electromagnetic radiation in the another frequency band.
2. The method of claim 1 wherein capturing an image includes separating the electromagnetic radiation into three mutually distinct frequency bands and capturing each frequency band with a different image sensor.
3. The method of claim 1 further comprises detecting an occlusion on the lens by quantifying the intensity of the electromagnetic radiation in each of the frequency bands, computing a difference in the intensity of the electromagnetic radiation between the frequency bands and tagging the image with an indicator of an occlusion in response to the difference being less than a threshold.
4. The method of claim 1 further comprises comparing the intensity of electromagnetic radiation by designating a plurality of bins for each image sensor, such that each bin represents a range of intensity values for electromagnetic radiation; assigning each pixel in a given image sensor into one of the plurality of bins; and counting, for a given image sensor, a number of pixels classified in each of the plurality of bins.
5. The method of claim 4 further comprises comparing pixel counts for corresponding bins amongst the different image sensors, determining how many of the corresponding bins have substantially same pixel count, and identifying an occlusion on the lens when the number of corresponding bins having substantially same pixel counts exceeds a threshold.
6. The method of claim 1 further comprises
constructing a clean model for an image captured by the camera without an occlusion;
constructing additional models for an image captured by the camera, where each additional model is for an image captured by the camera having a different type of occlusion thereon;
classifying, by machine learning, the image using the clean model and the additional models.
7. A system for detecting occlusions on a color camera, comprising:
a camera configured to capture an image of a scene, where electromagnetic radiation captured by the camera is captured in at least two mutually distinct frequency bands; and
an image processor in data communication with the camera, wherein the image processor analyzes intensity of electromagnetic radiation by comparing the intensity of electromagnetic radiation in one frequency band to the intensity of electromagnetic radiation in another frequency band, where the one frequency band is distinct from the another frequency band and the electromagnetic radiation for the one frequency band is taken from same image as the electromagnetic radiation for the another frequency band, and detects an occlusion of the lens of the camera when the intensity of electromagnetic radiation in the one frequency band is similar to the intensity of electromagnetic radiation in the another frequency band.
8. The system of claim 7 wherein the camera is further defined as a color camera.
9. The system of claim 7 wherein the image processor compares the intensity of electromagnetic radiation by designating a plurality of bins for each image sensor, such that each bin represents a range of intensity values for electromagnetic radiation; assigning each pixel in a given image sensor into one of the plurality of bins; and counting, for a given image sensor, a number of pixels classified in each of the plurality of bins.
10. The system of claim 9 wherein the image processor compares pixel counts for corresponding bins amongst the different image sensors, determines how many of the corresponding bins have substantially same pixel count, and identifies an occlusion on the lens when the number of corresponding bins having substantially same pixel counts exceeds a threshold.
11. A method for detecting occlusions on a color camera, comprising:
capturing, by the camera, an image of a scene, where electromagnetic radiation captured by the camera is captured in at least two mutually distinct frequency bands;
comparing, by an image processor, intensity of electromagnetic radiation in one frequency band to intensity of electromagnetic radiation in another frequency band by designating a plurality of bins for each image sensor, such that each bin represents a range of intensity values for electromagnetic radiation; assigning each pixel in a given image sensor into one of the plurality of bins; and counting, for a given image sensor, a number of pixels classified in each of the plurality of bins; and
detecting, by the image processor, an occlusion on a lens of the camera when the intensity of the electromagnetic radiation in the one frequency band is similar to the intensity of the electromagnetic radiation in the another frequency band.
12. The method of claim 11 further comprises comparing pixel counts for corresponding bins amongst the different image sensors, determining how many of the corresponding bins have substantially same pixel count, and identifying an occlusion on the lens when the number of corresponding bins having substantially same pixel counts exceeds a threshold.
13. The method of claim 11 further comprises detecting an occlusion on the lens by quantifying the intensity of the electromagnetic radiation in each of the frequency bands, computing a difference in the intensity of the electromagnetic radiation between the frequency bands and tagging the image with an indicator of an occlusion in response to the difference being less than a threshold.
14. The method of claim 11 further comprises
constructing a clean model for an image captured by the camera without an occlusion;
constructing additional models for an image captured by the camera, where each additional model is for an image captured by the camera having a different type of occlusion thereon;
classifying, by machine learning, the image using the clean model and the additional models.