subtracker/tablell.py at master · subotto/subtracker · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
#!/usr/bin/env python2
# -*- coding: utf-8 -*-

import cv2
import numpy

class FrameTableAnalysis:

    def __init__(self):
        # Log-likelihood of each pixel to be coming from the table background
        self.background_ll = None
        # Log-likelihood of each pixel to be coming from the ball
        self.ball_ll = None

        self.ballness = None


# The following anlyzer uses the variance matrix, but it still has to
# be tested. We use a simpler version instead.
def analyze_table_background_full(bg_estimation, warped_frame, controls):
    analysis = FrameTableAnalysis()

    # Compute multivariate normal PDF (see
    # wikipedia:Multivariate normal distribution).  We are exploiting
    # numpy broadcasting rules to do it pixel-by-pixel.

    h, w, channels = warped_frame.shape
    k = channels * 1.0

    x = warped_frame
    mu = bg_estimation.mean
    sigma = bg_estimation.variance

    term3 = -1 / 2.0 * numpy.einsum("...i,...ij,...j", x - mu, numpy.linalg.inv(sigma), x - mu)
    density = -k / 2 * numpy.log(2 * numpy.pi) - 1 / 2.0 * numpy.log(numpy.linalg.det(sigma)) + term3

    analysis.background_ll = density

    controls.show("background_ll", analysis.background_ll)

    return analysis

# Simpler version that uses the squared deviation normalized by a constant
def analyze_table_background(bg_estimation, warped_frame, controls):
    analysis = FrameTableAnalysis()

    deviation = (warped_frame - bg_estimation.mean)

    bg_sigma = controls.trackbar("sigma", 0.2, 0.01, 0.5, 0.01).value
    analysis.background_ll = numpy.sum(deviation * deviation, 2) * (-0.5 / bg_sigma ** 2)

    controls.show("background_ll", analysis.background_ll)

    ball_intensity = controls.trackbar("ball intensity", 0.95).value
    ball_sigma = controls.trackbar("ball sigma", 0.15).value

    ball_deviation = warped_frame - ball_intensity
    analysis.ball_ll = numpy.sum(ball_deviation * ball_deviation, 2) * (-0.5 / ball_sigma ** 2)

    controls.show("ball_ll", analysis.ball_ll)

    combined_ll = analysis.ball_ll - analysis.background_ll

    analysis.ballness = cv2.blur(combined_ll, (3, 3))
    controls.show("ballness", analysis.ballness)

    return analysis