File size: 14,259 Bytes
47af768
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
import os
import numpy as np


def iou_batch(bboxes1, bboxes2):
    """
    From SORT: Computes IOU between two bboxes in the form [x1,y1,x2,y2]
    """
    bboxes2 = np.expand_dims(bboxes2, 0)
    bboxes1 = np.expand_dims(bboxes1, 1)
    
    xx1 = np.maximum(bboxes1[..., 0], bboxes2[..., 0])
    yy1 = np.maximum(bboxes1[..., 1], bboxes2[..., 1])
    xx2 = np.minimum(bboxes1[..., 2], bboxes2[..., 2])
    yy2 = np.minimum(bboxes1[..., 3], bboxes2[..., 3])
    w = np.maximum(0., xx2 - xx1)
    h = np.maximum(0., yy2 - yy1)
    wh = w * h
    o = wh / ((bboxes1[..., 2] - bboxes1[..., 0]) * (bboxes1[..., 3] - bboxes1[..., 1])                                      
        + (bboxes2[..., 2] - bboxes2[..., 0]) * (bboxes2[..., 3] - bboxes2[..., 1]) - wh)                                              
    return(o)  


def giou_batch(bboxes1, bboxes2):
    """
    :param bbox_p: predict of bbox(N,4)(x1,y1,x2,y2)
    :param bbox_g: groundtruth of bbox(N,4)(x1,y1,x2,y2)
    :return:
    """
    # for details should go to https://arxiv.org/pdf/1902.09630.pdf
    # ensure predict's bbox form
    bboxes2 = np.expand_dims(bboxes2, 0)
    bboxes1 = np.expand_dims(bboxes1, 1)

    xx1 = np.maximum(bboxes1[..., 0], bboxes2[..., 0])
    yy1 = np.maximum(bboxes1[..., 1], bboxes2[..., 1])
    xx2 = np.minimum(bboxes1[..., 2], bboxes2[..., 2])
    yy2 = np.minimum(bboxes1[..., 3], bboxes2[..., 3])
    w = np.maximum(0., xx2 - xx1)
    h = np.maximum(0., yy2 - yy1)
    wh = w * h
    iou = wh / ((bboxes1[..., 2] - bboxes1[..., 0]) * (bboxes1[..., 3] - bboxes1[..., 1])                                      
        + (bboxes2[..., 2] - bboxes2[..., 0]) * (bboxes2[..., 3] - bboxes2[..., 1]) - wh)  

    xxc1 = np.minimum(bboxes1[..., 0], bboxes2[..., 0])
    yyc1 = np.minimum(bboxes1[..., 1], bboxes2[..., 1])
    xxc2 = np.maximum(bboxes1[..., 2], bboxes2[..., 2])
    yyc2 = np.maximum(bboxes1[..., 3], bboxes2[..., 3])
    wc = xxc2 - xxc1 
    hc = yyc2 - yyc1 
    assert((wc > 0).all() and (hc > 0).all())
    area_enclose = wc * hc 
    giou = iou - (area_enclose - wh) / area_enclose
    giou = (giou + 1.)/2.0 # resize from (-1,1) to (0,1)
    return giou


def diou_batch(bboxes1, bboxes2):
    """
    :param bbox_p: predict of bbox(N,4)(x1,y1,x2,y2)
    :param bbox_g: groundtruth of bbox(N,4)(x1,y1,x2,y2)
    :return:
    """
    # for details should go to https://arxiv.org/pdf/1902.09630.pdf
    # ensure predict's bbox form
    bboxes2 = np.expand_dims(bboxes2, 0)
    bboxes1 = np.expand_dims(bboxes1, 1)

    # calculate the intersection box
    xx1 = np.maximum(bboxes1[..., 0], bboxes2[..., 0])
    yy1 = np.maximum(bboxes1[..., 1], bboxes2[..., 1])
    xx2 = np.minimum(bboxes1[..., 2], bboxes2[..., 2])
    yy2 = np.minimum(bboxes1[..., 3], bboxes2[..., 3])
    w = np.maximum(0., xx2 - xx1)
    h = np.maximum(0., yy2 - yy1)
    wh = w * h
    iou = wh / ((bboxes1[..., 2] - bboxes1[..., 0]) * (bboxes1[..., 3] - bboxes1[..., 1])                                      
        + (bboxes2[..., 2] - bboxes2[..., 0]) * (bboxes2[..., 3] - bboxes2[..., 1]) - wh) 

    centerx1 = (bboxes1[..., 0] + bboxes1[..., 2]) / 2.0
    centery1 = (bboxes1[..., 1] + bboxes1[..., 3]) / 2.0
    centerx2 = (bboxes2[..., 0] + bboxes2[..., 2]) / 2.0
    centery2 = (bboxes2[..., 1] + bboxes2[..., 3]) / 2.0

    inner_diag = (centerx1 - centerx2) ** 2 + (centery1 - centery2) ** 2

    xxc1 = np.minimum(bboxes1[..., 0], bboxes2[..., 0])
    yyc1 = np.minimum(bboxes1[..., 1], bboxes2[..., 1])
    xxc2 = np.maximum(bboxes1[..., 2], bboxes2[..., 2])
    yyc2 = np.maximum(bboxes1[..., 3], bboxes2[..., 3])

    outer_diag = (xxc2 - xxc1) ** 2 + (yyc2 - yyc1) ** 2
    diou = iou - inner_diag / outer_diag

    return (diou + 1) / 2.0 # resize from (-1,1) to (0,1)

def ciou_batch(bboxes1, bboxes2):
    """
    :param bbox_p: predict of bbox(N,4)(x1,y1,x2,y2)
    :param bbox_g: groundtruth of bbox(N,4)(x1,y1,x2,y2)
    :return:
    """
    # for details should go to https://arxiv.org/pdf/1902.09630.pdf
    # ensure predict's bbox form
    bboxes2 = np.expand_dims(bboxes2, 0)
    bboxes1 = np.expand_dims(bboxes1, 1)

    # calculate the intersection box
    xx1 = np.maximum(bboxes1[..., 0], bboxes2[..., 0])
    yy1 = np.maximum(bboxes1[..., 1], bboxes2[..., 1])
    xx2 = np.minimum(bboxes1[..., 2], bboxes2[..., 2])
    yy2 = np.minimum(bboxes1[..., 3], bboxes2[..., 3])
    w = np.maximum(0., xx2 - xx1)
    h = np.maximum(0., yy2 - yy1)
    wh = w * h
    iou = wh / ((bboxes1[..., 2] - bboxes1[..., 0]) * (bboxes1[..., 3] - bboxes1[..., 1])                                      
        + (bboxes2[..., 2] - bboxes2[..., 0]) * (bboxes2[..., 3] - bboxes2[..., 1]) - wh) 

    centerx1 = (bboxes1[..., 0] + bboxes1[..., 2]) / 2.0
    centery1 = (bboxes1[..., 1] + bboxes1[..., 3]) / 2.0
    centerx2 = (bboxes2[..., 0] + bboxes2[..., 2]) / 2.0
    centery2 = (bboxes2[..., 1] + bboxes2[..., 3]) / 2.0

    inner_diag = (centerx1 - centerx2) ** 2 + (centery1 - centery2) ** 2

    xxc1 = np.minimum(bboxes1[..., 0], bboxes2[..., 0])
    yyc1 = np.minimum(bboxes1[..., 1], bboxes2[..., 1])
    xxc2 = np.maximum(bboxes1[..., 2], bboxes2[..., 2])
    yyc2 = np.maximum(bboxes1[..., 3], bboxes2[..., 3])

    outer_diag = (xxc2 - xxc1) ** 2 + (yyc2 - yyc1) ** 2
    
    w1 = bboxes1[..., 2] - bboxes1[..., 0]
    h1 = bboxes1[..., 3] - bboxes1[..., 1]
    w2 = bboxes2[..., 2] - bboxes2[..., 0]
    h2 = bboxes2[..., 3] - bboxes2[..., 1]

    # prevent dividing over zero. add one pixel shift
    h2 = h2 + 1.
    h1 = h1 + 1.
    arctan = np.arctan(w2/h2) - np.arctan(w1/h1)
    v = (4 / (np.pi ** 2)) * (arctan ** 2)
    S = 1 - iou 
    alpha = v / (S+v)
    ciou = iou - inner_diag / outer_diag - alpha * v
    
    return (ciou + 1) / 2.0 # resize from (-1,1) to (0,1)


def ct_dist(bboxes1, bboxes2):
    """
        Measure the center distance between two sets of bounding boxes,
        this is a coarse implementation, we don't recommend using it only
        for association, which can be unstable and sensitive to frame rate
        and object speed.
    """
    bboxes2 = np.expand_dims(bboxes2, 0)
    bboxes1 = np.expand_dims(bboxes1, 1)

    centerx1 = (bboxes1[..., 0] + bboxes1[..., 2]) / 2.0
    centery1 = (bboxes1[..., 1] + bboxes1[..., 3]) / 2.0
    centerx2 = (bboxes2[..., 0] + bboxes2[..., 2]) / 2.0
    centery2 = (bboxes2[..., 1] + bboxes2[..., 3]) / 2.0

    ct_dist2 = (centerx1 - centerx2) ** 2 + (centery1 - centery2) ** 2

    ct_dist = np.sqrt(ct_dist2)

    # The linear rescaling is a naive version and needs more study
    ct_dist = ct_dist / ct_dist.max()
    return ct_dist.max() - ct_dist # resize to (0,1)



def speed_direction_batch(dets, tracks):
    tracks = tracks[..., np.newaxis]
    CX1, CY1 = (dets[:,0] + dets[:,2])/2.0, (dets[:,1]+dets[:,3])/2.0
    CX2, CY2 = (tracks[:,0] + tracks[:,2]) /2.0, (tracks[:,1]+tracks[:,3])/2.0
    dx = CX1 - CX2 
    dy = CY1 - CY2 
    norm = np.sqrt(dx**2 + dy**2) + 1e-6
    dx = dx / norm 
    dy = dy / norm
    return dy, dx # size: num_track x num_det


def linear_assignment(cost_matrix):
    try:
        import lap
        _, x, y = lap.lapjv(cost_matrix, extend_cost=True)
        return np.array([[y[i],i] for i in x if i >= 0]) #
    except ImportError:
        from scipy.optimize import linear_sum_assignment
        x, y = linear_sum_assignment(cost_matrix)
        return np.array(list(zip(x, y)))


def associate_detections_to_trackers(detections,trackers,iou_threshold = 0.3):
    """
    Assigns detections to tracked object (both represented as bounding boxes)
    Returns 3 lists of matches, unmatched_detections and unmatched_trackers
    """
    if(len(trackers)==0):
        return np.empty((0,2),dtype=int), np.arange(len(detections)), np.empty((0,5),dtype=int)

    iou_matrix = iou_batch(detections, trackers)

    if min(iou_matrix.shape) > 0:
        a = (iou_matrix > iou_threshold).astype(np.int32)
        if a.sum(1).max() == 1 and a.sum(0).max() == 1:
            matched_indices = np.stack(np.where(a), axis=1)
        else:
            matched_indices = linear_assignment(-iou_matrix)
    else:
        matched_indices = np.empty(shape=(0,2))

    unmatched_detections = []
    for d, det in enumerate(detections):
        if(d not in matched_indices[:,0]):
            unmatched_detections.append(d)
    unmatched_trackers = []
    for t, trk in enumerate(trackers):
        if(t not in matched_indices[:,1]):
            unmatched_trackers.append(t)

    #filter out matched with low IOU
    matches = []
    for m in matched_indices:
        if(iou_matrix[m[0], m[1]]<iou_threshold):
            unmatched_detections.append(m[0])
            unmatched_trackers.append(m[1])
        else:
            matches.append(m.reshape(1,2))
    if(len(matches)==0):
        matches = np.empty((0,2),dtype=int)
    else:
        matches = np.concatenate(matches,axis=0)

    return matches, np.array(unmatched_detections), np.array(unmatched_trackers)


def associate(detections, trackers, iou_threshold, velocities, previous_obs, vdc_weight):    
    if(len(trackers)==0):
        return np.empty((0,2),dtype=int), np.arange(len(detections)), np.empty((0,5),dtype=int)

    Y, X = speed_direction_batch(detections, previous_obs)
    inertia_Y, inertia_X = velocities[:,0], velocities[:,1]
    inertia_Y = np.repeat(inertia_Y[:, np.newaxis], Y.shape[1], axis=1)
    inertia_X = np.repeat(inertia_X[:, np.newaxis], X.shape[1], axis=1)
    diff_angle_cos = inertia_X * X + inertia_Y * Y
    diff_angle_cos = np.clip(diff_angle_cos, a_min=-1, a_max=1)
    diff_angle = np.arccos(diff_angle_cos)
    diff_angle = (np.pi /2.0 - np.abs(diff_angle)) / np.pi

    valid_mask = np.ones(previous_obs.shape[0])
    valid_mask[np.where(previous_obs[:,4]<0)] = 0
    
    iou_matrix = iou_batch(detections, trackers)
    scores = np.repeat(detections[:,-1][:, np.newaxis], trackers.shape[0], axis=1)
    # iou_matrix = iou_matrix * scores # a trick sometiems works, we don't encourage this
    valid_mask = np.repeat(valid_mask[:, np.newaxis], X.shape[1], axis=1)

    angle_diff_cost = (valid_mask * diff_angle) * vdc_weight
    angle_diff_cost = angle_diff_cost.T
    angle_diff_cost = angle_diff_cost * scores

    if min(iou_matrix.shape) > 0:
        a = (iou_matrix > iou_threshold).astype(np.int32)
        if a.sum(1).max() == 1 and a.sum(0).max() == 1:
            matched_indices = np.stack(np.where(a), axis=1)
        else:
            matched_indices = linear_assignment(-(iou_matrix+angle_diff_cost))
    else:
        matched_indices = np.empty(shape=(0,2))

    unmatched_detections = []
    for d, det in enumerate(detections):
        if(d not in matched_indices[:,0]):
            unmatched_detections.append(d)
    unmatched_trackers = []
    for t, trk in enumerate(trackers):
        if(t not in matched_indices[:,1]):
            unmatched_trackers.append(t)

    # filter out matched with low IOU
    matches = []
    for m in matched_indices:
        if(iou_matrix[m[0], m[1]]<iou_threshold):
            unmatched_detections.append(m[0])
            unmatched_trackers.append(m[1])
        else:
            matches.append(m.reshape(1,2))
    if(len(matches)==0):
        matches = np.empty((0,2),dtype=int)
    else:
        matches = np.concatenate(matches,axis=0)

    return matches, np.array(unmatched_detections), np.array(unmatched_trackers)


def associate_kitti(detections, trackers, det_cates, iou_threshold, 
        velocities, previous_obs, vdc_weight):
    if(len(trackers)==0):
        return np.empty((0,2),dtype=int), np.arange(len(detections)), np.empty((0,5),dtype=int)

    """
        Cost from the velocity direction consistency
    """
    Y, X = speed_direction_batch(detections, previous_obs)
    inertia_Y, inertia_X = velocities[:,0], velocities[:,1]
    inertia_Y = np.repeat(inertia_Y[:, np.newaxis], Y.shape[1], axis=1)
    inertia_X = np.repeat(inertia_X[:, np.newaxis], X.shape[1], axis=1)
    diff_angle_cos = inertia_X * X + inertia_Y * Y
    diff_angle_cos = np.clip(diff_angle_cos, a_min=-1, a_max=1)
    diff_angle = np.arccos(diff_angle_cos)
    diff_angle = (np.pi /2.0 - np.abs(diff_angle)) / np.pi

    valid_mask = np.ones(previous_obs.shape[0])
    valid_mask[np.where(previous_obs[:,4]<0)]=0  
    valid_mask = np.repeat(valid_mask[:, np.newaxis], X.shape[1], axis=1)

    scores = np.repeat(detections[:,-1][:, np.newaxis], trackers.shape[0], axis=1)
    angle_diff_cost = (valid_mask * diff_angle) * vdc_weight
    angle_diff_cost = angle_diff_cost.T
    angle_diff_cost = angle_diff_cost * scores

    """
        Cost from IoU
    """
    iou_matrix = iou_batch(detections, trackers)
    

    """
        With multiple categories, generate the cost for catgory mismatch
    """
    num_dets = detections.shape[0]
    num_trk = trackers.shape[0]
    cate_matrix = np.zeros((num_dets, num_trk))
    for i in range(num_dets):
            for j in range(num_trk):
                if det_cates[i] != trackers[j, 4]:
                        cate_matrix[i][j] = -1e6
    
    cost_matrix = - iou_matrix -angle_diff_cost - cate_matrix

    if min(iou_matrix.shape) > 0:
        a = (iou_matrix > iou_threshold).astype(np.int32)
        if a.sum(1).max() == 1 and a.sum(0).max() == 1:
            matched_indices = np.stack(np.where(a), axis=1)
        else:
            matched_indices = linear_assignment(cost_matrix)
    else:
        matched_indices = np.empty(shape=(0,2))

    unmatched_detections = []
    for d, det in enumerate(detections):
        if(d not in matched_indices[:,0]):
            unmatched_detections.append(d)
    unmatched_trackers = []
    for t, trk in enumerate(trackers):
        if(t not in matched_indices[:,1]):
            unmatched_trackers.append(t)

    #filter out matched with low IOU
    matches = []
    for m in matched_indices:
        if(iou_matrix[m[0], m[1]]<iou_threshold):
            unmatched_detections.append(m[0])
            unmatched_trackers.append(m[1])
        else:
            matches.append(m.reshape(1,2))
    if(len(matches)==0):
        matches = np.empty((0,2),dtype=int)
    else:
        matches = np.concatenate(matches,axis=0)

    return matches, np.array(unmatched_detections), np.array(unmatched_trackers)