diff --git a/.idea/.gitignore b/.idea/.gitignore
new file mode 100644
index 0000000..10b731c
--- /dev/null
+++ b/.idea/.gitignore
@@ -0,0 +1,5 @@
+# 默认忽略的文件
+/shelf/
+/workspace.xml
+# 基于编辑器的 HTTP 客户端请求
+/httpRequests/
diff --git a/.idea/inspectionProfiles/Project_Default.xml b/.idea/inspectionProfiles/Project_Default.xml
new file mode 100644
index 0000000..fd02ef8
--- /dev/null
+++ b/.idea/inspectionProfiles/Project_Default.xml
@@ -0,0 +1,26 @@
+
+
+
+
+
+
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/.idea/inspectionProfiles/profiles_settings.xml b/.idea/inspectionProfiles/profiles_settings.xml
new file mode 100644
index 0000000..105ce2d
--- /dev/null
+++ b/.idea/inspectionProfiles/profiles_settings.xml
@@ -0,0 +1,6 @@
+
+
+
+
+
+
\ No newline at end of file
diff --git a/.idea/misc.xml b/.idea/misc.xml
new file mode 100644
index 0000000..dc5f227
--- /dev/null
+++ b/.idea/misc.xml
@@ -0,0 +1,7 @@
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/.idea/modules.xml b/.idea/modules.xml
new file mode 100644
index 0000000..a3e7011
--- /dev/null
+++ b/.idea/modules.xml
@@ -0,0 +1,8 @@
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/.idea/opi-yolo.iml b/.idea/opi-yolo.iml
new file mode 100644
index 0000000..c506ebe
--- /dev/null
+++ b/.idea/opi-yolo.iml
@@ -0,0 +1,12 @@
+
+
+
+
+
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/.idea/vcs.xml b/.idea/vcs.xml
new file mode 100644
index 0000000..8306744
--- /dev/null
+++ b/.idea/vcs.xml
@@ -0,0 +1,7 @@
+
+
+
+
+
+
+
\ No newline at end of file
diff --git a/infer.py b/infer.py
index 7b51b10..f0cfd40 100644
--- a/infer.py
+++ b/infer.py
@@ -1,78 +1,198 @@
+import os
import cv2
import numpy as np
+import argparse
+from ais_bench.infer.interface import InferSession
-# 导入昇腾核心推理库（板子自带）
-from acllite_model import AclLiteModel
-from acllite_resource import AclLiteResource
+class YOLO26Infer:
+ def __init__(self, model_path, device_id=0):
+ """
+ 初始化推理器
+ :param model_path: .om 模型的本地路径
+ :param device_id: NPU 设备ID
+ """
+ self.device_id = device_id
+ self.model_path = model_path
+ # 创建 ais_bench 推理会话
+ self.session = InferSession(device_id, model_path)
+ # 获取模型输入尺寸，用于后续预处理
+ self.input_width = 640
+ self.input_height = 640
-def preprocess(img_path, model_w=640, model_h=640):
- """图片预处理：缩放、通道转换、归一化"""
- orig_img = cv2.imread(img_path)
- h, w, _ = orig_img.shape
-
- # 1. 缩放到模型要求的 640x640
- img = cv2.resize(orig_img, (model_w, model_h))
- # 2. BGR 转 RGB (YOLO 训练时通常是 RGB)
- img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
- # 3. 归一化并调整维度为 [1, 3, 640, 640]
- img = img.astype(np.float32) / 255.0
- img = img.transpose(2, 0, 1) # HWC -> CHW
- img = np.expand_dims(img, axis=0) # CHW -> NCHW
-
- # 必须将其转换为连续的内存，方便 NPU 读取
- return np.ascontiguousarray(img), orig_img
+ def letterbox(self, img, new_shape=(640, 640), color=(114, 114, 114)):
+ shape = img.shape[:2]
+ r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
+ # 计算缩放后的新尺寸（四舍五入）
+ new_unpad = (int(round(shape[1] * r)), int(round(shape[0] * r)))
+ # 计算需要填充的总宽度和高度
+ dw = new_shape[1] - new_unpad[0]
+ dh = new_shape[0] - new_unpad[1]
+ # 分别计算左右/上下填充量（允许奇数不对称）
+ left = dw // 2
+ right = dw - left
+ top = dh // 2
+ bottom = dh - top
-def main():
- # 1. 初始化昇腾 NPU 资源
- acl_resource = AclLiteResource()
- acl_resource.init()
+ # 缩放图片
+ if shape[::-1] != new_unpad:
+ img = cv2.resize(img, new_unpad, interpolation=cv2.INTER_LINEAR)
- # 2. 加载你的 YOLO26 .om 模型
- model_path = "./yolo26_8t.om"
- model = AclLiteModel(model_path)
+ # 添加填充（可能左右或上下不对称，但总尺寸正好是 new_shape）
+ img = cv2.copyMakeBorder(img, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)
+ return img, r, (left, top) # 返回填充的左上角偏移量
- # 3. 图像预处理
- input_data, orig_img = preprocess("./test.jpg")
+ def preprocess(self, image_path):
+ """
+ 预处理图片，使其符合模型输入要求
+ """
+ # 使用OpenCV读取图片 (BGR格式)
+ img_bgr = cv2.imread(image_path)
+ if img_bgr is None:
+ raise FileNotFoundError(f"图片未找到: {image_path}")
- # 4. 送入 NPU 进行硬件加速推理
- # execute 接收一个列表，里面是模型的输入 Tensor
- result_list = model.execute([input_data])
+ # 获取原始图片尺寸，用于后处理时的坐标转换
+ self.original_height, self.original_width = img_bgr.shape[:2]
- # 5. 后处理（针对 YOLO26 的端到端特性）
- # YOLO26 输出的直接是最终预测框，result_list[0] 形状通常是 [1, 300, 6] 或类似
- # 6个维度的含义一般为：[x_min, y_min, x_max, y_max, score, class_id]
- predictions = result_list[0][0]
+ # 将图片从BGR转换为RGB
+ img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
- CONF_THRESHOLD = 0.25 # 设定置信度阈值
- orig_h, orig_w = orig_img.shape[:2]
+ # 调整图片尺寸并保持宽高比，进行Letterbox填充
+ img_resized, self.ratio, (self.pad_left, self.pad_top) = self.letterbox(
+ img_rgb, (self.input_width, self.input_height)
+ )
- box_count = 0
- for pred in predictions:
- # 取出置信度和类别
- score = pred[4]
- class_id = int(pred[5])
-
- # 过滤掉低置信度的框
- if score > CONF_THRESHOLD:
- box_count += 1
- # 取出缩放后的坐标 [0~640]
- x1, y1, x2, y2 = pred[0:4]
-
- # 将坐标还原到原图尺寸
- x1 = int(x1 * orig_w / 640)
- y1 = int(y1 * orig_h / 640)
- x2 = int(x2 * orig_w / 640)
- y2 = int(y2 * orig_h / 640)
-
- # 在原图上画框和标签
- cv2.rectangle(orig_img, (x1, y1), (x2, y2), (0, 255, 0), 2)
- cv2.putText(orig_img, f"Class {class_id}: {score:.2f}", (x1, y1 - 10),
- cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
+ # 归一化：将像素值从 [0, 255] 转为 [0, 1]
+ img_input = (img_resized.astype(np.float32) / 255.0).astype(np.float32)
- print(f"推理完成，共检测到 {box_count} 个目标。")
-
- # 6. 保存检测结果图
- cv2.imwrite("./result.jpg", orig_img)
+ # 调整维度：HWC -> NCHW，并转换为连续数组
+ img_input = np.ascontiguousarray(np.transpose(img_input, (2, 0, 1))[np.newaxis, :, :, :])
+
+ return img_input
+
+ def infer(self, img_input):
+ """
+ 执行推理
+ """
+ # 使用会话的 infer 方法进行推理
+ # infer 方法接收一个列表，包含输入张量，返回一个列表，包含输出张量
+ result = self.session.infer([img_input])
+ print(f"Input tensor shape: {img_input.shape}")
+ print(f"Input tensor dtype: {img_input.dtype}")
+ print(f"Input tensor size: {img_input.nbytes}")
+ return result
+
+ def postprocess(self, infer_result, conf_threshold=0.5):
+ """
+ 对模型输出进行后处理
+ :param infer_result: 模型原始输出，一个列表，通常第一个元素是输出张量
+ :param conf_threshold: 置信度阈值
+ :return: 包含检测结果的列表，每个元素为 [x1, y1, x2, y2, confidence, class_id]
+ """
+ detections = []
+ # 获取输出张量，假设输出列表的第一个元素是输出数组
+ if isinstance(infer_result, (list, tuple)):
+ output = infer_result[0]
+ else:
+ output = infer_result
+
+ # ========== 调试信息 ==========
+ # print(f"[DEBUG] output shape: {output.shape}")
+ # print(f"[DEBUG] output dtype: {output.dtype}")
+ # print(f"[DEBUG] output min: {output.min()}, max: {output.max()}")
+ # # 打印前5个检测框（如果形状是 [N,6] 或 [1,N,6]）
+ # if output.ndim == 3:
+ # flat_output = output[0]
+ # else:
+ # flat_output = output
+ # print(f"[DEBUG] first 5 rows:\n{flat_output[:5]}")
+
+ # 假设输出维度为 [batch, num_detections, 6]
+ # 移除batch维度，并只取第一个batch的结果
+ if output.ndim == 3:
+ output = output[0]
+
+ # 遍历所有检测框
+ for det in output:
+ confidence = det[4]
+ if confidence < conf_threshold:
+ continue
+
+ # 获取原始输出坐标 (x1, y1, x2, y2) 和 类别ID
+ x1_scale, y1_scale, x2_scale, y2_scale = det[0], det[1], det[2], det[3]
+ cls_id = int(det[5])
+
+ # 将坐标从模型输入尺寸映射回原始图片尺寸
+ x1 = (x1_scale - self.pad_left) / self.ratio
+ y1 = (y1_scale - self.pad_top) / self.ratio
+ x2 = (x2_scale - self.pad_left) / self.ratio
+ y2 = (y2_scale - self.pad_top) / self.ratio
+
+ # 边界限制，防止超出原始图片范围
+ x1 = max(0, min(x1, self.original_width))
+ y1 = max(0, min(y1, self.original_height))
+ x2 = max(0, min(x2, self.original_width))
+ y2 = max(0, min(y2, self.original_height))
+
+ detections.append([x1, y1, x2, y2, confidence, cls_id])
+
+ return detections
+
+def draw_boxes(image, detections, class_names=None):
+ """
+ 在图片上绘制检测框和标签
+ :param image: 原始图片数组 (BGR格式)
+ :param detections: 后处理得到的结果列表
+ :param class_names: 类别名称列表，默认使用类别ID
+ """
+ img_copy = image.copy()
+ for det in detections:
+ x1, y1, x2, y2, conf, cls_id = det
+ x1, y1, x2, y2 = int(x1), int(y1), int(x2), int(y2)
+
+ # 绘制矩形框
+ cv2.rectangle(img_copy, (x1, y1), (x2, y2), (0, 255, 0), 2)
+
+ # 构建标签文本
+ label = f"Class: {cls_id}" if class_names is None else class_names[cls_id]
+ label += f" {conf:.2f}"
+
+ # 设置标签背景
+ (label_w, label_h), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)
+ cv2.rectangle(img_copy, (x1, y1 - label_h - 5), (x1 + label_w, y1), (0, 255, 0), -1)
+ cv2.putText(img_copy, label, (x1, y1 - 5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 1)
+
+ return img_copy
if __name__ == "__main__":
- main()
\ No newline at end of file
+ parser = argparse.ArgumentParser(description="YOLOv26 目标检测推理")
+ parser.add_argument("--model", type=str, required=True, help="YOLOv26 OM模型文件路径")
+ parser.add_argument("--image", type=str, required=True, help="输入图片路径")
+ parser.add_argument("--conf", type=float, default=0.5, help="置信度阈值")
+ args = parser.parse_args()
+
+ # 初始化推理器
+ yolo_infer = YOLO26Infer(args.model)
+
+ # 图片预处理
+ print("预处理图片...")
+ input_tensor = yolo_infer.preprocess(args.image)
+
+ # 执行推理
+ print("执行推理...")
+ raw_result = yolo_infer.infer(input_tensor)
+
+ # 后处理
+ print("后处理...")
+ detections = yolo_infer.postprocess(raw_result, conf_threshold=args.conf)
+
+ # 打印结果
+ print(f"检测到 {len(detections)} 个目标:")
+ for det in detections:
+ print(f"类别ID: {int(det[5])}, 置信度: {det[4]:.4f}, 边界框: ({det[0]:.2f}, {det[1]:.2f}, {det[2]:.2f}, {det[3]:.2f})")
+
+ # 可选：绘制结果并保存
+ original_img = cv2.imread(args.image)
+ result_img = draw_boxes(original_img, detections)
+ output_path = "output_" + os.path.basename(args.image)
+ cv2.imwrite(output_path, result_img)
+ print(f"结果图片已保存至: {output_path}")
\ No newline at end of file
diff --git a/mav-test.py b/mav-test.py
new file mode 100644
index 0000000..1c570e4
--- /dev/null
+++ b/mav-test.py
@@ -0,0 +1,175 @@
+import cv2
+import math
+import time
+import serial
+import numpy as np
+from ais_bench.infer.interface import InferSession
+from pymavlink import mavutil
+
+
+class DroneDetector:
+ def __init__(self, model_path, serial_port='/dev/ttyAMA0', baudrate=57600):
+ self.session = InferSession(0, model_path)
+ self.input_size = 640
+ # 串口/MAVLink
+ self.serial_port = serial_port
+ self.baudrate = baudrate
+ self.mav = None
+ self.init_serial()
+
+ def init_serial(self):
+ """初始化串口（如果用于自定义协议）或 MAVLink 连接"""
+ try:
+ self.mav = mavutil.mavlink_connection(self.serial_port, baud=self.baudrate)
+ print(f"MAVLink 已连接到 {self.serial_port} 波特率 {self.baudrate}")
+ except Exception as e:
+ print(f"串口打开失败: {e}")
+ self.mav = None
+
+ def letterbox(self, img, color=(114, 114, 114)):
+ shape = img.shape[:2]
+ r = min(self.input_size / shape[0], self.input_size / shape[1])
+ new_unpad = (int(round(shape[1] * r)), int(round(shape[0] * r)))
+ dw = self.input_size - new_unpad[0]
+ dh = self.input_size - new_unpad[1]
+ left = dw // 2
+ right = dw - left
+ top = dh // 2
+ bottom = dh - top
+ img = cv2.resize(img, new_unpad, interpolation=cv2.INTER_LINEAR)
+ img = cv2.copyMakeBorder(img, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)
+ return img, r, (left, top)
+
+ def preprocess(self, img):
+ img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
+ img_resized, self.ratio, (self.pad_left, self.pad_top) = self.letterbox(img_rgb)
+ img_input = img_resized.astype(np.float32) / 255.0
+ img_input = np.ascontiguousarray(np.transpose(img_input, (2, 0, 1))[np.newaxis, ...])
+ return img_input
+
+ def postprocess(self, output, conf_thresh=0.5):
+ detections = []
+ if output.ndim == 3:
+ output = output[0]
+ for det in output:
+ raw_conf = det[4]
+ conf = 1.0 / (1.0 + math.exp(-raw_conf))
+ if conf < conf_thresh:
+ continue
+ x1, y1, x2, y2 = det[0], det[1], det[2], det[3]
+ # 映射回原图坐标
+ x1 = (x1 - self.pad_left) / self.ratio
+ y1 = (y1 - self.pad_top) / self.ratio
+ x2 = (x2 - self.pad_left) / self.ratio
+ y2 = (y2 - self.pad_top) / self.ratio
+ cls_id = int(det[5])
+ detections.append({
+ 'bbox': (x1, y1, x2, y2),
+ 'confidence': conf,
+ 'class': cls_id,
+ 'center': ((x1 + x2) / 2, (y1 + y2) / 2)
+ })
+ return detections
+
+ def send_mavlink_obstacle(self, angle_deg, distance_m):
+ """发送 DISTANCE_SENSOR 消息，表示目标方向与距离"""
+ if not self.mav:
+ return
+ self.mav.mav.distance_sensor_send(
+ 0, # time_boot_ms
+ 0, # min_distance cm
+ 10000, # max_distance cm
+ int(distance_m * 100), # current_distance cm
+ 0, # type (0=laser)
+ 0, # id
+ int(angle_deg / 360.0 * 65535), # orientation (0~360° -> 0~65535)
+ 0 # covariance
+ )
+
+ def send_custom_serial(self, detections):
+ """通过串口发送自定义格式的数据（例如 JSON）"""
+ if not self.serial or not self.serial.is_open:
+ return
+ import json
+ data = {
+ 'timestamp': time.time(),
+ 'detections': detections
+ }
+ self.serial.write((json.dumps(data) + '\n').encode())
+
+ def detect_and_send(self, frame, send_method='mavlink', conf_thresh=0.5):
+ """对单帧图像进行检测，并发送结果"""
+ input_tensor = self.preprocess(frame)
+ outputs = self.session.infer([input_tensor])
+ detections = self.postprocess(outputs[0], conf_thresh)
+ if not detections:
+ return detections
+ # 假设只发送置信度最高的那个目标
+ best = detections[0]
+ if send_method == 'mavlink':
+ # 计算角度和距离（需要根据相机参数和实际尺寸估算）
+ # 这里简单示例：根据目标在图像中的偏移量估算角度
+ cx = best['center'][0]
+ img_w = frame.shape[1]
+ angle = (cx - img_w / 2) / (img_w / 2) * 45 # 水平视场角90°，偏移量映射为±45°
+ # 距离估算：根据目标在图像中的高度（假设目标实际高度1m）
+ # 需要相机焦距等，简化：使用固定值10米
+ distance = 10.0
+ self.send_mavlink_obstacle(angle, distance)
+ elif send_method == 'custom':
+ self.send_custom_serial(detections)
+ return detections
+
+ def run_on_camera(self, camera_id=0, conf_thresh=0.5, send_method='mavlink'):
+ cap = cv2.VideoCapture(camera_id)
+ if not cap.isOpened():
+ print("摄像头打开失败")
+ return
+ print("开始检测，按 q 退出")
+ while True:
+ ret, frame = cap.read()
+ if not ret:
+ break
+ detections = self.detect_and_send(frame, send_method, conf_thresh)
+ # 绘制检测框（调试）
+ for det in detections:
+ x1, y1, x2, y2 = map(int, det['bbox'])
+ cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
+ cv2.putText(frame, f"{det['class']}:{det['confidence']:.2f}", (x1, y1 - 5),
+ cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
+ cv2.imshow("Detection", frame)
+ if cv2.waitKey(1) & 0xFF == ord('q'):
+ break
+ cap.release()
+ cv2.destroyAllWindows()
+
+ def run_on_image(self, image_path, conf_thresh=0.5, send_method='mavlink'):
+ frame = cv2.imread(image_path)
+ if frame is None:
+ print("图片读取失败")
+ return
+ detections = self.detect_and_send(frame, send_method, conf_thresh)
+ print(f"检测到 {len(detections)} 个目标")
+ for d in detections:
+ print(d)
+ # 显示结果
+ for det in detections:
+ x1, y1, x2, y2 = map(int, det['bbox'])
+ cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
+ cv2.putText(frame, f"{det['class']}:{det['confidence']:.2f}", (x1, y1 - 5),
+ cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
+ cv2.imshow("Detection", frame)
+ cv2.waitKey(0)
+ cv2.destroyAllWindows()
+
+
+if __name__ == "__main__":
+ MODEL_PATH = "./om/yolo26n.om"
+ SERIAL_PORT = "/dev/ttyAMA0" # 根据实际修改
+ BAUD = 57600
+
+ detector = DroneDetector(MODEL_PATH, SERIAL_PORT, BAUD)
+
+ # 选择模式：摄像头或单张图片
+ # detector.run_on_camera(camera_id=0, conf_thresh=0.5, send_method='mavlink')
+ detector.run_on_image("test.jpg", conf_thresh=0.5, send_method='mavlink')
\ No newline at end of file
diff --git a/test.jpg b/test.jpg
new file mode 100644
index 0000000..200912c
Binary files /dev/null and b/test.jpg differ