Tracking Under Low-light Conditions Using Background Subtraction Matthew Bennink Clemson University Clemson, SC
Outline ► Introduction ► Methods Camera Calibration Background Subtraction ► Experimental Results Uniform Light Single Light No Light ► Conclusions
Introduction ► We want to track objects with little or no light present. This will allow greater flexibility. ► Others have used infrared cameras and achieved good results. ► We will try to track objects using low-light cameras. These cameras have a set of LEDs around the lens to add ambient light to the environment. ► We will approach the problem by producing an occupancy map of the area using background subtraction.
Camera Calibration ► Camera calibration involves producing a mapping of image coordinates to world coordinates. ► Matlab toolbox used to calibrate cameras.
Background Subtraction Thresholding used to reduce noise. Very fast allowing for real-time image processing. Borrowed from Dr. Birchfield ’ s notes
Algorithm ► Calibrate cameras ► Create lookup table ► Store background images ► Create mask images ► Loop over time OccMap[x,y] = 1 For each camera n over all pixels (x,y) ► D[n,x,y] = |I[n,x,y]-B[n,x,y]| > T ? 1 : 0 ► If D[n,x,y] == 1 OccMap[Lookup[n,x,y]] = 0 Display OccMap
Experimental Results (Full Light) The tracking system performs fairly well. Principal Components Analysis (PCA) could improve this image.
Experimental Results (Single Light) Shadows cause problems. Current research is ongoing to remove noise from shadows.
Experimental Results (No Light) Poor results with little or no light present. Smoothing may reduce noise. Infrared sensors may offer solution.
Conclusions ► Use of background subtraction with low- light cameras is not the optimal solution ► Infrared sensors possibly provide better results.
Thank You! Questions ??