diff --git a/config/config.yaml b/config/config.yaml index 7b1d3a7..96cb058 100644 --- a/config/config.yaml +++ b/config/config.yaml @@ -11,7 +11,7 @@ model: distance_estimation: focal_length: 1000 # 摄像头焦距 - sensor_height: 5.6 # 摄像头传感器高度(mm) + sensor_height: 3000 # 摄像头传感器高度(mm) average_person_height: 1700 # 平均人身高(mm) api: diff --git a/docs/传感器尺寸.txt b/docs/传感器尺寸.txt new file mode 100644 index 0000000..8eb2248 --- /dev/null +++ b/docs/传感器尺寸.txt @@ -0,0 +1,84 @@ +def get_sensor_size_from_camera(): + """从相机API获取传感器尺寸""" + try: + import v4l2 + import fcntl + + # 打开相机设备 + fd = open('/dev/video0', 'rb') + + # 获取相机信息 + cp = v4l2.v4l2_capability() + fcntl.ioctl(fd, v4l2.VIDIOC_QUERYCAP, cp) + + # 获取相机格式 + fmt = v4l2.v4l2_format() + fmt.type = v4l2.V4L2_BUF_TYPE_VIDEO_CAPTURE + fcntl.ioctl(fd, v4l2.VIDIOC_G_FMT, fmt) + + # 获取传感器信息 + sensor_info = v4l2.v4l2_sensor_info() + fcntl.ioctl(fd, v4l2.VIDIOC_G_SENSOR_INFO, sensor_info) + + return { + 'width_mm': sensor_info.physical_width / 1000, # 转换为mm + 'height_mm': sensor_info.physical_height / 1000 + } + except Exception as e: + print(f"无法从相机API获取传感器尺寸: {e}") + return None + +def calibrate_sensor_size(real_object_size_mm, object_distance_mm): + """通过已知物体标定传感器尺寸""" + def capture_calibration_image(): + # 捕获标定图像 + cap = cv2.VideoCapture(0) + ret, frame = cap.read() + cap.release() + return frame + + def measure_object_pixels(frame): + # 测量物体在图像中的像素尺寸 + gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY) + edges = cv2.Canny(gray, 50, 150) + contours, _ = cv2.findContours(edges, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) + + if contours: + largest_contour = max(contours, key=cv2.contourArea) + x, y, w, h = cv2.boundingRect(largest_contour) + return h # 返回物体高度（像素） + return None + + # 捕获标定图像 + frame = capture_calibration_image() + if frame is None: + return None + + # 测量物体像素尺寸 + object_pixels = measure_object_pixels(frame) + if object_pixels is None: + return None + + # 使用相似三角形原理计算传感器尺寸 + frame_height = frame.shape[0] # 图像高度（像素） + focal_length_mm = 35 # 假设焦距为35mm，需要根据实际相机参数调整 + + sensor_height_mm = (real_object_size_mm * focal_length_mm) / (object_distance_mm * object_pixels / frame_height) + + return sensor_height_mm + +SENSOR_SIZES = { + '1/4"': {'width_mm': 3.2, 'height_mm': 2.4}, + '1/3"': {'width_mm': 4.8, 'height_mm': 3.6}, + '1/2.3"': {'width_mm': 6.17, 'height_mm': 4.55}, + '1/1.8"': {'width_mm': 7.18, 'height_mm': 5.32}, + '2/3"': {'width_mm': 8.8, 'height_mm': 6.6}, + '1"': {'width_mm': 13.2, 'height_mm': 8.8}, + '4/3"': {'width_mm': 17.3, 'height_mm': 13}, + 'APS-C': {'width_mm': 23.6, 'height_mm': 15.6}, + 'Full Frame': {'width_mm': 36, 'height_mm': 24} +} + +def get_sensor_size(sensor_format): + """根据传感器格式获取尺寸""" + return SENSOR_SIZES.get(sensor_format, None) diff --git a/output/detection_20250113_110502_763716.jpg b/output/detection_20250113_110502_763716.jpg deleted file mode 100644 index 8237005..0000000 Binary files a/output/detection_20250113_110502_763716.jpg and /dev/null differ diff --git a/output/detection_20250113_112407_136399.jpg b/output/detection_20250113_112407_136399.jpg deleted file mode 100644 index 4cb6ae2..0000000 Binary files a/output/detection_20250113_112407_136399.jpg and /dev/null differ diff --git a/output/detection_20250113_141239_166263.jpg b/output/detection_20250113_141239_166263.jpg new file mode 100644 index 0000000..47ec38c Binary files /dev/null and b/output/detection_20250113_141239_166263.jpg differ diff --git a/output/detection_20250113_155844_316627.jpg b/output/detection_20250113_155844_316627.jpg new file mode 100644 index 0000000..47ec38c Binary files /dev/null and b/output/detection_20250113_155844_316627.jpg differ diff --git a/src/__pycache__/distance_estimator.cpython-39.pyc b/src/__pycache__/distance_estimator.cpython-39.pyc index 1df163c..a3bcd68 100644 Binary files a/src/__pycache__/distance_estimator.cpython-39.pyc and b/src/__pycache__/distance_estimator.cpython-39.pyc differ diff --git a/src/distance_estimator.py b/src/distance_estimator.py index 558bd72..d36af44 100644 --- a/src/distance_estimator.py +++ b/src/distance_estimator.py @@ -1,15 +1,177 @@ import numpy as np +import cv2 +import mediapipe as mp class DistanceEstimator: - def __init__(self, focal_length, sensor_height, avg_person_height): - self.focal_length = focal_length - self.sensor_height = sensor_height - self.avg_person_height = avg_person_height + def __init__(self, + focal_length_mm, # 焦距(mm) + sensor_width_mm, # 传感器宽度(mm) + sensor_height_mm, # 传感器高度(mm) + image_width_pixels, # 图像宽度(像素) + image_height_pixels, # 图像高度(像素) + camera_height_mm=1700,# 摄像头安装高度(mm) + camera_tilt_angle=0 # 摄像头俯仰角(度) + ): + # 相机参数 + self.focal_length_mm = focal_length_mm + self.sensor_width_mm = sensor_width_mm + self.sensor_height_mm = sensor_height_mm + self.image_width_pixels = image_width_pixels + self.image_height_pixels = image_height_pixels + self.camera_height_mm = camera_height_mm + self.camera_tilt_angle = np.radians(camera_tilt_angle) - def estimate_distance(self, person_height_pixels, frame_height): - # 使用相似三角形原理估算距离 - real_height = self.avg_person_height # mm - sensor_height_pixels = frame_height + # 计算像素尺寸 + self.pixel_size_mm = sensor_width_mm / image_width_pixels - distance = (real_height * self.focal_length) / (person_height_pixels * self.sensor_height / sensor_height_pixels) - return distance # 返回距离(mm) \ No newline at end of file + # 计算水平和垂直视场角 + self.fov_horizontal = 2 * np.arctan(sensor_width_mm / (2 * focal_length_mm)) + self.fov_vertical = 2 * np.arctan(sensor_height_mm / (2 * focal_length_mm)) + + # 初始化人体关键点检测器 + self.mp_pose = mp.solutions.pose + self.pose = self.mp_pose.Pose( + static_image_mode=False, + model_complexity=2, + min_detection_confidence=0.5, + min_tracking_confidence=0.5 + ) + + # 人体参数（统计平均值，单位：mm） + self.body_ratios = { + 'eye_height_ratio': 0.936, # 眼睛高度占总高度比例 + 'shoulder_width_ratio': 0.259, # 肩宽占身高比例 + 'hip_width_ratio': 0.191, # 臀宽占身高比例 + 'avg_male_height': 1700, # 成年男性平均身高 + 'avg_female_height': 1600 # 成年女性平均身高 + } + + def estimate_distance_from_keypoints(self, frame): + """使用人体关键点进行距离估计""" + results = self.pose.process(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)) + if not results.pose_landmarks: + return None, None + + landmarks = results.pose_landmarks.landmark + + # 获取关键点 + left_shoulder = landmarks[self.mp_pose.PoseLandmark.LEFT_SHOULDER] + right_shoulder = landmarks[self.mp_pose.PoseLandmark.RIGHT_SHOULDER] + left_hip = landmarks[self.mp_pose.PoseLandmark.LEFT_HIP] + right_hip = landmarks[self.mp_pose.PoseLandmark.RIGHT_HIP] + + # 计算肩宽和臀宽（像素） + shoulder_width_px = np.sqrt( + (right_shoulder.x - left_shoulder.x)**2 * self.image_width_pixels**2 + + (right_shoulder.y - left_shoulder.y)**2 * self.image_height_pixels**2 + ) + + hip_width_px = np.sqrt( + (right_hip.x - left_hip.x)**2 * self.image_width_pixels**2 + + (right_hip.y - left_hip.y)**2 * self.image_height_pixels**2 + ) + + # 使用肩宽估算距离 + shoulder_distance = (self.body_ratios['shoulder_width_ratio'] * + self.body_ratios['avg_male_height'] * + self.focal_length_mm) / (shoulder_width_px * self.pixel_size_mm) + + # 使用臀宽估算距离 + hip_distance = (self.body_ratios['hip_width_ratio'] * + self.body_ratios['avg_male_height'] * + self.focal_length_mm) / (hip_width_px * self.pixel_size_mm) + + # 加权平均两个距离估计 + estimated_distance = 0.6 * shoulder_distance + 0.4 * hip_distance + + # 计算置信度（基于关键点可见性） + confidence = (left_shoulder.visibility + right_shoulder.visibility + + left_hip.visibility + right_hip.visibility) / 4 + + return estimated_distance, confidence + + def estimate_distance_from_bbox(self, bbox_height_px, estimated_height_mm=None): + """使用边界框进行距离估计""" + if estimated_height_mm is None: + estimated_height_mm = self.body_ratios['avg_male_height'] + + # 使用相似三角形原理 + distance = (estimated_height_mm * self.focal_length_mm) / (bbox_height_px * self.pixel_size_mm) + return distance + + def estimate_distance_from_ground_plane(self, feet_y_px): + """使用地平面投影进行距离估计""" + # 计算从图像底部到脚部位置的像素距离 + bottom_to_feet_px = self.image_height_pixels - feet_y_px + + # 计算视角 + angle_to_feet = self.camera_tilt_angle + (bottom_to_feet_px / self.image_height_pixels) * self.fov_vertical + + # 使用三角函数计算距离 + distance = self.camera_height_mm / np.tan(angle_to_feet) + return distance + + def estimate_distance(self, frame, bbox): + """ + 综合多种方法估计距离 + + Args: + frame: 输入图像帧 + bbox: 人物边界框 (x1, y1, x2, y2) + + Returns: + distance_mm: 估计距离（毫米） + confidence: 置信度（0-1） + """ + distances = [] + weights = [] + + # 1. 使用关键点估计 + keypoint_distance, keypoint_conf = self.estimate_distance_from_keypoints(frame) + if keypoint_distance is not None: + distances.append(keypoint_distance) + weights.append(keypoint_conf * 0.4) # 关键点方法权重0.4 + + # 2. 使用边界框估计 + bbox_height = bbox[3] - bbox[1] + bbox_distance = self.estimate_distance_from_bbox(bbox_height) + distances.append(bbox_distance) + weights.append(0.3) # 边界框方法权重0.3 + + # 3. 使用地平面投影估计 + ground_distance = self.estimate_distance_from_ground_plane(bbox[3]) + distances.append(ground_distance) + weights.append(0.3) # 地平面投影方法权重0.3 + + # 加权平均得到最终距离 + weights = np.array(weights) / np.sum(weights) # 归一化权重 + final_distance = np.average(distances, weights=weights) + + # 计算置信度（基于估计值的一致性） + std_dev = np.std(distances) + mean_distance = np.mean(distances) + consistency = np.exp(-std_dev / mean_distance) # 估计值越一致，置信度越高 + + # 如果有关键点检测结果，将其纳入最终置信度 + if keypoint_distance is not None: + final_confidence = (consistency + keypoint_conf) / 2 + else: + final_confidence = consistency + + return final_distance, final_confidence + + def visualize_estimation(self, frame, bbox, distance_mm, confidence): + """可视化距离估计结果""" + # vis_frame = frame.copy() + vis_frame = frame + x1, y1, x2, y2 = map(int, bbox) + + # 绘制边界框 + cv2.rectangle(vis_frame, (x1, y1), (x2, y2), (0, 255, 0), 2) + + # 添加距离和置信度信息 + text = f"Distance: {distance_mm/1000:.2f}m Conf: {confidence:.2f}" + cv2.putText(vis_frame, text, (x1, y1-10), + cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2) + + return vis_frame \ No newline at end of file diff --git a/test.py b/test.py index 64bd988..5916d75 100644 --- a/test.py +++ b/test.py @@ -1,6 +1,6 @@ import asyncio from tests.run_tests import run_all_tests - +# 测试生成程序 if __name__ == "__main__": asyncio.run(run_all_tests()) \ No newline at end of file diff --git a/tests/__pycache__/run_tests.cpython-39.pyc b/tests/__pycache__/run_tests.cpython-39.pyc index a589eb8..881861f 100644 Binary files a/tests/__pycache__/run_tests.cpython-39.pyc and b/tests/__pycache__/run_tests.cpython-39.pyc differ diff --git a/tests/__pycache__/test_camera.cpython-39.pyc b/tests/__pycache__/test_camera.cpython-39.pyc index fd0b661..b8c5c31 100644 Binary files a/tests/__pycache__/test_camera.cpython-39.pyc and b/tests/__pycache__/test_camera.cpython-39.pyc differ diff --git a/tests/__pycache__/test_distance_estimator.cpython-39.pyc b/tests/__pycache__/test_distance_estimator.cpython-39.pyc index 241cdeb..02687ae 100644 Binary files a/tests/__pycache__/test_distance_estimator.cpython-39.pyc and b/tests/__pycache__/test_distance_estimator.cpython-39.pyc differ diff --git a/tests/run_tests.py b/tests/run_tests.py index 4e71850..913fff4 100644 --- a/tests/run_tests.py +++ b/tests/run_tests.py @@ -9,8 +9,8 @@ async def run_all_tests(): # print("\n1. 测试摄像头模块") # test_rtsp_camera() - print("\n2. 测试人物检测模块") - test_person_detector() + # print("\n2. 测试人物检测模块") + # test_person_detector() print("\n3. 测试距离估算模块") test_distance_estimator() diff --git a/tests/test_camera.py b/tests/test_camera.py index ac8791d..0ebc6a3 100644 --- a/tests/test_camera.py +++ b/tests/test_camera.py @@ -5,11 +5,11 @@ from src.camera_handler import RTSPCamera def test_rtsp_camera(): # 使用本地测试视频或摄像头 - camera = RTSPCamera(0) # 使用本地摄像头进行测试 + camera = RTSPCamera("rtsp://10.0.0.17:8554/camera_test/2") # 使用本地摄像头进行测试 print("测试摄像头启动...") camera.start() - time.sleep(2) # 等待摄像头初始化 + time.sleep(10) # 等待摄像头初始化 print("测试获取帧...") frame = camera.get_frame() diff --git a/tests/test_data/distance_estimation_result.jpg b/tests/test_data/distance_estimation_result.jpg new file mode 100644 index 0000000..34929d4 Binary files /dev/null and b/tests/test_data/distance_estimation_result.jpg differ diff --git a/tests/test_data/distance_estimation_result_1.jpg b/tests/test_data/distance_estimation_result_1.jpg new file mode 100644 index 0000000..f6eef25 Binary files /dev/null and b/tests/test_data/distance_estimation_result_1.jpg differ diff --git a/tests/test_data/2024-04-02-04_14_55-749386_jpg.rf.6e256601cc6513bf8e6f4497f4662d71.jpg b/tests/test_data/test_image_1.jpg similarity index 100% rename from tests/test_data/2024-04-02-04_14_55-749386_jpg.rf.6e256601cc6513bf8e6f4497f4662d71.jpg rename to tests/test_data/test_image_1.jpg diff --git a/tests/test_distance_estimator.py b/tests/test_distance_estimator.py index 5aa6788..ede88b6 100644 --- a/tests/test_distance_estimator.py +++ b/tests/test_distance_estimator.py @@ -1,31 +1,73 @@ +import cv2 import numpy as np from src.distance_estimator import DistanceEstimator +from src.person_detector import PersonDetector def test_distance_estimator(): - # 初始化距离估算器 - focal_length = 35 # mm - sensor_height = 24 # mm - avg_person_height = 1700 # mm - + # 初始化距离估算器（使用实际相机参数） estimator = DistanceEstimator( - focal_length=focal_length, - sensor_height=sensor_height, - avg_person_height=avg_person_height + focal_length_mm=35, + sensor_width_mm=23.5, + sensor_height_mm=15.6, + image_width_pixels=1920, + image_height_pixels=1080, + camera_height_mm=1700, + camera_tilt_angle=15 ) print("测试距离估算...") - # 测试场景1：人物占据画面一半高度 - frame_height = 1080 - person_height_pixels = frame_height / 2 - distance = estimator.estimate_distance(person_height_pixels, frame_height) - print(f"场景1 - 人物占据画面一半高度时的估算距离: {distance:.2f}mm") - assert distance > 0, "距离估算结果应该大于0" + # 读取测试图像 + test_image = cv2.imread("tests/test_data/test_image.jpg") + assert test_image is not None, "无法加载测试图像" + + # 初始化检测器 + detector = PersonDetector("yolov8n.pt", conf_threshold=0.3) + persons, vis_path = detector.detect(test_image, save_visualization=True) + + print("len(person)", len(persons)) + + for person in persons: + assert 'bbox' in person, "检测结果缺少bbox" + assert 'confidence' in person, "检测结果缺少confidence" + assert len(person['bbox']) == 4, "bbox格式不正确" + assert 0 <= person['confidence'] <= 1, "confidence值范围不正确" + + - # 测试场景2：人物占据画面四分之一高度 - person_height_pixels = frame_height / 4 - distance = estimator.estimate_distance(person_height_pixels, frame_height) - print(f"场景2 - 人物占据画面四分之一高度时的估算距离: {distance:.2f}mm") - assert distance > 0, "距离估算结果应该大于0" + # 测试边界框方法 + bbox = person['bbox'] # 示例边界框 + bbox_distance = estimator.estimate_distance_from_bbox(bbox[3] - bbox[1]) + print(f"边界框方法估算距离: {bbox_distance:.2f}mm") + assert bbox_distance > 0, "距离估算结果应该大于0" + + # 测试关键点方法 + keypoint_distance, keypoint_conf = estimator.estimate_distance_from_keypoints(test_image) + if keypoint_distance is not None: + print(f"关键点方法估算距离: {keypoint_distance:.2f}mm, 置信度: {keypoint_conf:.2f}") + assert keypoint_distance > 0, "距离估算结果应该大于0" + assert 0 <= keypoint_conf <= 1, "置信度应该在0-1之间" + + # 测试地平面投影方法 + ground_distance = estimator.estimate_distance_from_ground_plane(bbox[3]) + print(f"地平面投影方法估算距离: {ground_distance:.2f}mm") + assert ground_distance > 0, "距离估算结果应该大于0" + + # 测试综合估计方法 + final_distance, final_confidence = estimator.estimate_distance(test_image, bbox) + print(f"综合估计距离: {final_distance:.2f}mm, 置信度: {final_confidence:.2f}") + assert final_distance > 0, "距离估算结果应该大于0" + assert 0 <= final_confidence <= 1, "置信度应该在0-1之间" + + # 测试可视化 + vis_image = estimator.visualize_estimation(test_image, bbox, final_distance, final_confidence) + assert vis_image is not None, "可视化结果不应为空" + assert vis_image.shape == test_image.shape, "可视化图像尺寸应与原图相同" + + # 保存可视化结果 + cv2.imwrite("tests/test_data/distance_estimation_result.jpg", vis_image) - print("距离估算测试完成!") \ No newline at end of file + print("距离估算测试完成!") + +if __name__ == "__main__": + test_distance_estimator() \ No newline at end of file