5.24

lane_server/infer.py

@@ -0,0 +1,25 @@
+import paddle.inference as paddle_infer
+import numpy as np
+import paddle.vision.transforms as T
+
+class Lane_model_infer:
+    def __init__(self, model_dir="./lane_model"):
+        # 初始化paddle推理
+        self.model_dir = model_dir
+        self.config = paddle_infer.Config(model_dir + "/model.pdmodel", model_dir + "/model.pdiparams")
+        self.config.enable_use_gpu(200, 0)
+        self.predictor = paddle_infer.create_predictor(self.config)
+        self.input_names = self.predictor.get_input_names()
+        self.input_handle = self.predictor.get_input_handle(self.input_names[0])
+        self.output_names = self.predictor.get_output_names()
+        self.output_handle = self.predictor.get_output_handle(self.output_names[0])
+        self.normalize_transform = T.Normalize(mean=[127.5], std=[127.5])
+    
+    def infer(self,src) -> np.ndarray:
+        image = self.normalize_transform(src)
+        image = image.transpose(2, 0, 1)
+        image = np.expand_dims(image, axis=0)
+        self.input_handle.copy_from_cpu(image)
+        self.predictor.run()
+        results = self.output_handle.copy_to_cpu()[0]
+        return results

lane_server/lane_infer_server.py

@@ -0,0 +1,59 @@
+import toml
+import threading
+from loguru import logger
+import zmq
+from infer import Lane_model_infer
+import numpy as np
+
+lock = threading.Lock()
+response = {'code': 0, 'data': 0}
+
+# 处理server响应数据
+def server_resp(lane_infer_port):
+    logger.info("lane server thread init success")
+    global response
+    
+    context = zmq.Context()
+    # 启动server
+    socket = context.socket(zmq.REP)
+    socket.bind(f"tcp://*:{lane_infer_port}")
+    logger.info("lane infer server init success")
+    while True:
+        message = socket.recv_string()
+        with lock:
+            socket.send_pyobj(response)
+
+
+
+if __name__ == "__main__":
+
+    cfg = toml.load('../cfg_infer_server.toml')
+
+    # 配置日志输出
+    logger.add(cfg['debug']['logger_filename'], format=cfg['debug']['logger_format'], retention = 5, level="INFO")
+    camera_cfg = cfg['camera']
+
+    # 连接摄像头server 巡线只需要连接前摄像头
+    context = zmq.Context()
+    camera_socket = context.socket(zmq.SUB)
+    camera_socket.connect(f"tcp://*:{camera_cfg['front_camera_port']}")
+    camera_socket.setsockopt(zmq.SUBSCRIBE, '')
+    logger.info("connect camera success")
+    # 初始化paddle推理器
+    predictor = Lane_model_infer()
+    logger.info("lane model load success")
+    # 启动lane_infer_server线程
+    mythread = threading.Thread(target=server_resp,
+                                args=(camera_cfg['lane_infer_port'],),
+                                daemon=True)
+    mythread.start()
+
+    while True:
+        image_data_bytes = camera_socket.recv(320*240)
+        image = np.frombuffer(image_data_bytes, dtype=np.uint8).reshape((240, 320, 3))
+        result = predictor.infer(image)
+        with lock:
+            response['data'] = result
+    mythread.join()
+    logger.info("lane infer server exit")
+