road_seg_statistics.cpp
15.6 KB
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
#include "road_seg_statistics.h"
#include "omp.h"
RoadSegProcess::RoadSegProcess(){
m_max_batchsize = 16;
}
RoadSegProcess::~RoadSegProcess(){
release();
}
cv::Mat RoadSegProcess::mask_to_rgb(cv::Mat img, cv::Mat mask) {
cv::Mat masks = img.clone();
int reg_cls = 8;
for (int i = 0; i < masks.rows; i++) {
for (int j = 0; j < masks.cols; j++) {
for (int k = 1; k < reg_cls; k++) {
if (mask.at<int>(i,j) == k) {
masks.at<cv::Vec3b>(i,j)[0] = seg_colors[k][0];
masks.at<cv::Vec3b>(i,j)[1] = seg_colors[k][1];
masks.at<cv::Vec3b>(i,j)[2] = seg_colors[k][2];
}
}
}
}
return masks;
}
float RoadSegProcess::contourArea(std::vector<cv::Point> contour, cv::Point2f& center) {
cv::RotatedRect rect = cv::minAreaRect(contour); // 最小外接矩形 rect[0]中心点 rect[1]宽 高 rect[2]旋转角度
center = rect.center;
return cv::contourArea(contour);
}
void RoadSegProcess::lanes_process(const rs_lane* lanes, int lane_count, std::vector<std::pair<std::vector<cv::Point>, int>>& combined, float scale_w, float scale_h) {
std::vector<std::vector<cv::Point> > lanes_xys;
std::vector<int> lanes_cls;
for (int i = 0; i < lane_count; i++) {
std::vector<cv::Point> xys;
for (int j = 0; j < lanes[i].num_points; j++) {
int x = static_cast<int>(lanes[i].points[j].x_ * scale_w);
int y = static_cast<int>(lanes[i].points[j].y_ * scale_h);
if (x > 0 && y > 0) {
xys.emplace_back(x, y);
}
}
if (!xys.empty()) {
lanes_xys.push_back(xys);
lanes_cls.push_back(lanes[i].cls);
}
}
for (size_t i = 0; i < lanes_xys.size(); ++i) {
combined.push_back(std::make_pair(lanes_xys[i], lanes_cls[i]));
}
if (!combined.empty()) {
//按车道线起点坐标排序,相应的类别顺序也会变化以保证标签对齐
std::sort(combined.begin(), combined.end(), [](const std::pair<std::vector<cv::Point>, int>& a, const std::pair<std::vector<cv::Point>, int>& b) {
return a.first[0].x < b.first[0].x;
});
}
}
cv::Mat RoadSegProcess::imshow_lanes(cv::Mat img, const rs_lane* lanes, int lane_count) {
float scale_w = img.cols / 640.0;
float scale_h = img.rows / 360.0;
std::vector<std::pair<std::vector<cv::Point>, int>> combined;
lanes_process(lanes, lane_count, combined, scale_w, scale_h);
for (const auto& lane_info : combined) {
const auto& xys = lane_info.first;
int cls = lane_info.second;
cv::Scalar color(lane_colors[cls][0],lane_colors[cls][1],lane_colors[cls][2]);
for (size_t i = 1; i < xys.size(); ++i) {
cv::line(img, xys[i - 1], xys[i], color, 4);
}
}
return img;
}
int RoadSegProcess::Mask2LanePoints(const cv::Mat& pred, std::vector<std::vector<cv::Point>>&lanes, std::vector<int>& cats) {
std::vector<int> labels = {8, 9, 10, 11};
for(auto cat: labels) {
cv::Mat b_masks, measure_labels, stats, centroids;
cv::compare(pred, cat, b_masks, cv::CMP_EQ); //将pred元素逐个和cat比较，相同255，不同0
// cv::threshold(pred, b_masks, 126, 255, cv::THRESH_OTSU); //生成二值图
// 连通域计算 b_masks:闭操作后的二值图像 measure_labels:和原图一样大的标记图 centroids:nccomps×2的矩阵,表示每个连通域的质心(x, y)
// stats:nccomps×5的矩阵 表示每个连通区域的外接矩形和面积(x, y, w, h, area)，索引0是背景信息
int nccomps = cv::connectedComponentsWithStats (b_masks, measure_labels, stats, centroids, 4); //8
for(int cv_measure_id = 1; cv_measure_id < nccomps; cv_measure_id++ ) { //跳过背景信息0
cv::Mat cv_measure_mask;
cv::compare(measure_labels, cv_measure_id, cv_measure_mask, cv::CMP_EQ);
std::vector< std::vector< cv::Point> > contours;
cv::findContours(cv_measure_mask,contours,cv::noArray(),cv::RETR_TREE,cv::CHAIN_APPROX_SIMPLE);
std::vector<cv::Point> contours_poly;
for (int j = 0; j < contours.size();j++) {
float area = cv::contourArea(contours[j]); if (area < 60) continue; //30
cv::approxPolyDP(cv::Mat(contours[j]), contours_poly, 6, true); //减小epsilon提高拟合精度(需要调试，epsilon过小会使多边形不够简化，单条线变多条)
if (contours_poly.size() == 1) continue;
lanes.push_back(contours_poly);
cats.push_back(cat);
}
}
}
return 0;
}
cv::Mat RoadSegProcess::seg_post_process(bool large_resolution, unsigned char *seg_array, std::vector<std::pair<std::vector<cv::Point>, int>> combined, std::vector<std::vector<cv::Point>> &poly_masks, std::vector<int> ®ion_classes, std::vector<std::vector<cv::Point>> &lanes, std::vector<int> &cats, std::map<double, int> &x_sort) {
std::vector<int> pred_cls;
int h = 360, w = 640;
cv::Mat lanes_masks = cv::Mat_<int>(h,w); //正常分割结果
cv::Mat mask_rmlane = cv::Mat_<int>(h,w); //车道线区域置为背景
cv::Mat solve_masks = cv::Mat_<int>(h,w); //计算主行驶区域用(虚线及减速标线归入行车道)
int step_size = h*w;
int seg_min_region_area = 512; //1024
for (int i = 0; i < h; ++i) {
for (int j = 0; j < w; ++j) {
int max_cls = seg_array[(i * w + j)];
lanes_masks.ptr<int>(i)[j] = max_cls;
pred_cls.push_back(max_cls);
mask_rmlane.ptr<int>(i)[j] = max_cls;
solve_masks.ptr<int>(i)[j] = max_cls;
}
}
for (const auto& lane_info : combined) {
const auto& xys = lane_info.first;
int cls = lane_info.second;
if (cls == 1) cls = 8; /*黄实线*/ if (cls == 2) cls = 9; /*白实线*/
if (cls == 3) cls = 10; /*虚线*/ if (cls == 4) cls = 11; /*黄虚线*/
for (size_t i = 1; i < xys.size(); ++i) {
cv::line(lanes_masks, xys[i - 1], xys[i], cls, 4); //绘制车道线用于求连通域
if (cls == 10 || cls == 11) cls = 1; //求主行驶区域时将虚线归入行车道
cv::line(solve_masks, xys[i - 1], xys[i], cls, 4);
}
}
/* 求背景区域--mask车道区域，场景变化用
cv::compare(lanes_masks, 0, background_mask, cv::CMP_EQ); //将lanes_masks元素逐个和0比较，相同255，不同0*/
#if 1
//--mask远处区域------------------------------------------------
cv::Mat mask_black = mask_rmlane.clone();
if (large_resolution) mask_black(cv::Rect(0, 0, w, int(h * 0.14))) = 0;
else mask_black(cv::Rect(0, 0, w, int(h * 0.22))) = 0;
mask_black(cv::Rect(0, int(h * 0.95), w, int(h * 0.05))) = 0;
mask_black(cv::Rect(0, 0, int(w * 0.02), h)) = 0;
mask_black(cv::Rect(int(w * 0.95), 0, int(w * 0.05), h)) = 0;
mask_rmlane = mask_black;
//-------------------------------------------------------------------------
#endif
//2.去重获取预测到的类别
std::vector<int> labels(pred_cls);
std::sort(labels.begin(),labels.end());
labels.erase(std::unique(labels.begin(),labels.end()),labels.end());
//4.求车道线区域
int flag = Mask2LanePoints(lanes_masks, lanes, cats); ///////////////////////// 车道线如何与mask结合
//5.求道路区域
int count = 0;
for(auto cat: labels) {
// std::cout << cat << std::endl;
cv::Mat b_masks, measure_labels, stats, centroids;
cv::Mat n_masks = cv::Mat_<int>(h,w);
n_masks = cv::Scalar(255);
cv::compare(mask_rmlane, cat, b_masks, cv::CMP_EQ); //将mask_rmlane元素逐个和cat比较，相同255，不同0
//连通域计算
int nccomps = cv::connectedComponentsWithStats (b_masks, measure_labels, stats, centroids, 8);
for(int i = 1; i < nccomps; i++ ) { //跳过背景信息0
//移除过小的区域，并将对应位置置为0
if (stats.at<int>(i, cv::CC_STAT_AREA) < seg_min_region_area) {
cv::Mat comparison_result;
cv::compare(measure_labels, cv::Scalar(i), comparison_result, cv::CMP_EQ); //相等为255不等为0
n_masks.setTo(0, comparison_result); // 将comparison_result中非零区域置为0
cv::bitwise_and(mask_rmlane,n_masks,mask_rmlane);
continue;
}
double centr_x = centroids.at<double>(i, 0);
double centr_y = centroids.at<double>(i, 1);
int region_class = cat;
// printf("region_class: %d\n", region_class);
cv::Mat region_mask;
cv::bitwise_and(measure_labels,n_masks,measure_labels);/////
cv::compare(measure_labels, i, region_mask, cv::CMP_EQ);
std::vector< std::vector< cv::Point> > contours;
cv::findContours(region_mask,contours,cv::noArray(),cv::RETR_TREE,cv::CHAIN_APPROX_SIMPLE);
std::vector<cv::Point> contours_poly;
for (int j = 0; j < contours.size();j++) {
cv::approxPolyDP(cv::Mat(contours[j]), contours_poly, 10, true);
if (contours_poly.size() <= 2) continue;
poly_masks.push_back(contours_poly);
region_classes.push_back(cat);
if (x_sort.count(centr_x)) centr_x += 0.0001;
x_sort.insert(std::make_pair(centr_x, count));
++ count;
// for (auto iter: contours_poly) {
// std::cout << "contour " << iter << " " << cat << std::endl;
// }
}
}
}
#if 0
//6.draw lanes
for (int i = 0; i < lanes.size(); ++i) {
int thickness = 4;
for (int j = 0; j < lanes[i].size()-1; ++j) {
cv::line(mask_rmlane, lanes[i][j], lanes[i][j+1], {seg_colors[cats[i]][0],seg_colors[cats[i]][1],seg_colors[cats[i]][2]}, thickness);
// std::cout << lanes[i][j] << " " << lanes[i][j+1] << " " << cats[i] << std::endl;
}
}
#endif
return mask_rmlane;
}
/* 算法初始化 */
int RoadSegProcess::init(int gpu_id, string models_dir) {
// string model_path = models_dir + "/models/village/tzroad_seg_310p.om" ;
string model_path = models_dir + "/models/village/tzroad_seg1027_310p.om" ;
LOG_INFO("road_seg 版本：{} 模型路径：{}", rs_get_version(), model_path);
rs_param param;
char modelNames[100];
strcpy(modelNames, model_path.c_str());
param.modelNames = modelNames;
param.thresld = 0.25;
param.devId = gpu_id;
m_devId = param.devId;
ACL_CALL(aclrtSetDevice(m_devId), ACL_SUCCESS, -1);
ACL_CALL(aclrtCreateContext(&m_algorthim_ctx, m_devId), ACL_SUCCESS, -1);
int ret = rs_init(&m_seg_handle, param);
if(ret != 0) {
LOG_ERROR("Init RoadSegSdk failed error code is {}", ret);
return -1;
}
return 0;
}
/* 算法计算 */
int RoadSegProcess::process_gpu(sy_img * batch_img, vector<DeviceMemory*> vec_segMem, vector<string>& tasklist,
vector<vector<vector<int>>>& traffic_region, vector<vector<int>>& labels)
{
int batchsize = tasklist.size();
traffic_region.resize(batchsize), labels.resize(batchsize);
rs_result result[batchsize];
int fea_size = SEG_IMG_RES_W * SEG_IMG_RES_H;
for (int b = 0; b < batchsize; b++) {
result[b].seg_array = new unsigned char[fea_size];
}
/* 路数太多时 按照最大batchsize数 拆批次运行 */
int ret = aclrtSetCurrentContext(m_algorthim_ctx);
if (ACL_SUCCESS != ret) {
return false;
}
int cur_batch_size = m_max_batchsize;
int cycleTimes = batchsize / cur_batch_size + (batchsize % cur_batch_size == 0 ? 0 : 1);
for (int c = 0; c < cycleTimes; c++) {
int real_batchsize = c == cycleTimes - 1 ? (batchsize - cur_batch_size*c) : cur_batch_size;
int startbatch = c*cur_batch_size;
rs_result *real_res = result + startbatch;
ret = rs_batch(m_seg_handle, batch_img + startbatch, real_batchsize, real_res);
}
for (int b = 0; b < batchsize; b++) {
std::vector<std::pair<std::vector<cv::Point>, int>> combined;
lanes_process(result[b].reg_array, result[b].lane_count, combined);
std::vector<std::vector<cv::Point>> poly_masks, lanes;
std::vector<int> region_classes, cats;
std::map<double, int> x_sort;
bool large_resolution = false;
if (batch_img[b].w_ > 1920) large_resolution = true;
cv::Mat seg_output = seg_post_process(large_resolution, result[b].seg_array, combined, poly_masks, region_classes, lanes, cats, x_sort); //m_masks:mask前的结果 poly_masks后的结果
float width_ratio = batch_img[b].w_*1.0 / SEG_IMG_RES_W;
float height_ratio = batch_img[b].h_*1.0 / SEG_IMG_RES_H;
for (int i = 0; i < lanes.size(); ++i) {
std::vector<cv::Point> points = lanes[i];
int lane_cls = cats[i];
vector<int> cur_region;
for (int j = 0; j < points.size(); ++j) {
int px = points[j].x*width_ratio;
int py = points[j].y*height_ratio;
cur_region.push_back(px); cur_region.push_back(py);
}
traffic_region[b].push_back(cur_region); // 存储区域
labels[b].push_back(lane_cls); // 存储类别
}
for (int i = 0; i < poly_masks.size(); ++i) {
std::vector<cv::Point> points = poly_masks[i];
int seg_cls = region_classes[i];
vector<int> cur_region;
for (int j = 0; j < points.size(); ++j) {
int px = points[j].x*width_ratio;
int py = points[j].y*height_ratio;
cur_region.push_back(px);
cur_region.push_back(py);
}
traffic_region[b].push_back(cur_region); // 存储区域
labels[b].push_back(seg_cls); // 存储类别
}
#if 0
{
jpegUtil.jpeg_init(m_devId);
std::string file_path = "/data/shzhao/vpt_ascend/bin/res/seg/";
auto time_now = std::chrono::system_clock::now();
std::string cur_timestamp_us = std::to_string(std::chrono::duration_cast<std::chrono::microseconds>(time_now.time_since_epoch()).count());
std::string img_filename = file_path + cur_timestamp_us + "_" + tasklist[b] + ".jpg";
vpc_img_info src_img_info = VPCUtil::vpc_devMem2vpcImg(vec_segMem[b]);
bool bSaved = jpegUtil.jpeg_encode(src_img_info.pic_desc, img_filename);
std::cout << bSaved << " " << img_filename << std::endl;
cv::Mat cur_img = cv::imread(img_filename);
cv::resize(cur_img, cur_img, cv::Size(640,360), 0, 0, cv::INTER_LINEAR);
cv::Mat vis_image = mask_to_rgb(cur_img, seg_output);
cv::imwrite(img_filename, vis_image);
VPCUtil::vpc_img_release(src_img_info);
jpegUtil.jpeg_release();
}
#endif
std::vector<std::vector<cv::Point>>().swap(poly_masks);
std::vector<std::vector<cv::Point>>().swap(lanes);
std::vector<int>().swap(region_classes);
std::vector<int>().swap(cats);
}
for (int b = 0; b < batchsize; b++) {
delete[] result[b].seg_array; result[b].seg_array = NULL;
}
return 0;
}
/* 算法句柄 资源释放 */
void RoadSegProcess::release(){
if (m_seg_handle){
rs_release(&m_seg_handle);
m_seg_handle = NULL;
}
if(m_algorthim_ctx){
aclrtDestroyContext(m_algorthim_ctx);
}
}