re-id/bbox_trigger.py at master · PurdueVIP-HumanBehavior/re-id · 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
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
from skimage.measure import compare_ssim
import cv2
from utils import crop_image
import numpy as np
import collections


class BboxTrigger:
    """
    This class is bounding box trigger for a door to build the gallery

    Attributes:
    _camera_id (str): name of the camera location
    _ref_img (cv2 image): reference image (ex. Closed door)
    _open_thresh (int): threshold value for opening a door
    _close_thresh (int): threshold value for closing a door
    _check_coords (list): 2D list of coordinates to check
    _sample_coords (list): 2D list of trigger coordinates
    _detector (detector.py): object detector (default FasterRCNN)
    _check (int): flag for determining when triggering is in process
    """
    def __init__(self, camera_id, ref_img, open_thresh, close_thresh,
                 check_coords, sample_coords, detector):
        """
        the constructor for BboxTrigger class

        Parameters:
        _camera_id (str): name of the camera location
        _ref_img (cv2 image): reference image (ex. Closed door)
        _open_thresh (int): threshold value for opening a door
        _close_thresh (int): threshold value for closing a door
        _check_coords (list): 2D list of coordinates to check
        _sample_coords (list): 2D list of trigger coordinates
        _detector (detector.py): object detector (default FasterRCNN)

        """
        self._camera_id = camera_id
        self._open_thresh = open_thresh
        self._close_thresh = close_thresh
        self._check_coords = check_coords
        self._sample_coords = sample_coords
        self._detector = detector  # ideally this is not here in the future either
        self._ref_img = cv2.cvtColor(crop_image(ref_img, check_coords),
                                     cv2.COLOR_BGR2GRAY)

        self._check = 0

    def update(self, frames):
        """
        Given a image, find when to trigger and return bounding boxes of people in the trigger region

        Parameters:
        frames (dict): dictionary of frames from all the cameras

        Returns:
        bboxes (ndarry): bounding boxes of detected objects
        sampimg (ndarray): cropped image of the trigger region
        """
        img = frames[self._camera_id]
        chkimg = cv2.cvtColor(crop_image(img, self._check_coords),
                              cv2.COLOR_BGR2GRAY)
        score, diff = compare_ssim(self._ref_img, chkimg, full=True)

        if self._check == 1:
            if score > self._close_thresh:
                sampimg = crop_image(img, self._sample_coords)
                bboxes, scores = self._detector.get_bboxes(sampimg)
                self._check = 0
                return True, bboxes, sampimg
        else:
            if score < self._open_thresh:
                self._check = 1

        return False, None, None

"use is completely different from BboxTrigger"
class VectorTrigger:
    """
    This class is for a Line trigger


    """
    def __init__(self, video, vector, inpt, length_thresh, frame_offset):
        tmpvec = np.random.randn(2)
        maxx = max(vector[2], vector[0])
        minx = min(vector[2], vector[0])
        maxy = max(vector[3], vector[1])
        miny = min(vector[3], vector[1])
        invec = np.array([maxx - minx, maxy - miny])
        # tmpvec -= tmpvec.dot(invec) * invec
        # print(tmpvec.dot(invec))
        # self.ovector = tmpvec / np.linalg.norm(tmpvec)
        self.ovector = np.array([invec[1], -1 * invec[0]])
        self.midpt = np.array([(vector[0] + vector[2])/2, (vector[1] + vector[3])/2])
        # 1 is in
        if np.sign(self.ovector.dot(self.midpt - inpt)) != 1: self.ovector = np.zeros(2) - self.ovector
        self.length_thresh = length_thresh
        self.frame_offset = frame_offset

        self.video_oi = video
        self.flags = collections.defaultdict(float)
        self.prev_val = collections.defaultdict(float)

    def update(self, peoplebboxes):
        """
        :param peoplebboxes: numpy array n x 5 with columns in x1, y1, x2, y2, id order
        :return: indexes to capture
        """
        retval = list()
        feetpoints = np.array([(peoplebboxes[:,0] + peoplebboxes[:,2])/2, np.max(peoplebboxes[:, [1,3]], 1)]).transpose()
        displace_vects = self.midpt - feetpoints
        disp_mags = np.linalg.norm(displace_vects, axis=1)

        inout = np.sign(displace_vects.dot(self.ovector))

        for i, (val, bboxes) in enumerate(zip(inout, peoplebboxes)):
            if disp_mags[i] > self.length_thresh:
                continue
            id = bboxes[4]

            if self.flags[id] >= 1: self.flags[id] += 1
            if self.flags[id] >= self.frame_offset:
                retval.append(i)
                self.flags[id] = 0

            # -1 - 1 is entering
            if self.prev_val[id] - val == -2:
                self.flags[id] = 1

            if val == 0:
                self.prev_val[id] = -1
            else: self.prev_val[id] = val

        return retval