OpenVINO + OpenCV实现车辆检测与道路分割
小白学视觉
共 2285字,需浏览 5分钟
·
2020-11-01 20:11
转载:OpenCV学堂
模型介绍
OpenVINO支持道路分割与车辆检测,预训练模型分别为:
- road-segmentation-adas-0001
- vehicle-detection-adas-0002
其中道路分割模型的输出四个分类,格式如下:
BG, road, curb, mark, 输出格式[NCHW]=[1x4x512x896]
车辆检测模型基于SSD MobileNetv1实现,输出格式为:
NCHW = [1x1xNx7],其中N表示检测到boxes数目
代码演示
01
道路分割模型加载与推理
首先加载道路分割模型,代码如下:
1# 道路分割
2net = ie.read_network(model=model_xml, weights=model_bin)
3input_blob = next(iter(net.input_info))
4out_blob = next(iter(net.outputs))
5
6n, c, h, w = net.input_info[input_blob].input_data.shape
7print(n, c, h, w)
8
9cap = cv.VideoCapture("D:/images/video/project_video.mp4")
10exec_net = ie.load_network(network=net, device_name="CPU")
推理与解析
1# 推理道路分割
2image = cv.resize(frame, (w, h))
3image = image.transpose(2, 0, 1)
4res = exec_net.infer(inputs={input_blob: [image]})
5
6# 解析道路分割结果
7res = res[out_blob]
8res = np.squeeze(res, 0)
9res = res.transpose(1, 2, 0)
10res = np.argmax(res, 2)
11print(res.shape)
12hh, ww = res.shape
13mask = np.zeros((hh, ww, 3), dtype=np.uint8)
14mask[np.where(res > 0)] = (0, 255, 255)
15mask[np.where(res > 1)] = (255, 0, 255)
16mask = cv.resize(mask, (frame.shape[1], frame.shape[0]))
17result = cv.addWeighted(frame, 0.5, mask, 0.5, 0)
02
车辆检测模型加载与推理解析
加载车辆检测模型,推理与解析SSD输出结果的代码如下:
1# 车辆检测
2vnet = ie.read_network(model=vehicel_xml, weights=vehicel_bin)
3vehicle_input_blob = next(iter(vnet.input_info))
4vehicle_out_blob = next(iter(vnet.outputs))
5
6vn, vc, vh, vw = vnet.input_info[vehicle_input_blob].input_data.shape
7print(n, c, h, w)
8vehicle_exec_net = ie.load_network(network=vnet, device_name="CPU")
9
10# 车辆检测
11inf_start = time.time()
12image = cv.resize(frame, (vw, vh))
13image = image.transpose(2, 0, 1)
14vec_res = vehicle_exec_net.infer(inputs={vehicle_input_blob:[image]})
15
16# 解析车辆检测结果
17ih, iw, ic = frame.shape
18vec_res = vec_res[vehicle_out_blob]
19for obj in vec_res[0][0]:
20 if obj[2] > 0.5:
21 xmin = int(obj[3] * iw)
22 ymin = int(obj[4] * ih)
23 xmax = int(obj[5] * iw)
24 ymax = int(obj[6] * ih)
25 cv.rectangle(frame, (xmin, ymin), (xmax, ymax), (0, 0, 255), 2, 8)
26 cv.putText(frame, str(obj[2]), (xmin, ymin), cv.FONT_HERSHEY_PLAIN, 1.0, (0, 0, 255), 1)
运行结果如下:
视频文件
评论