🏆 来源：第四届3D视觉创新应用竞赛 | 🧑💻 作者：马元豪（河南工业大学） 📦 仓库：Skeleton_Posture-recognition
🧠 项目概述
还在用示教器一点点摇机械臂？太 Out 啦！本项目打造了一套“所见即所得”的具身智能交互系统，只需动动你的手，机械臂就能精准跟随并执行抓取/释放动作。
系统巧妙地结合了2D手部关键点检测与3D深度感知，实现了手部空间坐标的精准定位；同时，采用静态手势（握拳/张开）与动态手势（轨迹/动作）相结合的方式，让机械臂既懂“去哪儿”，又懂“干什么”。
💡 核心亮点
– 端边协同架构：Jetson Orin Nano 负责视觉感知（轻量级），PC 负责动态识别与控制（算力密集型），算力分配恰到好处。
– 3D 视觉闭环：基于 Gemini2 深度相机，将 2D 像素坐标瞬间映射为 3D 空间坐标，直接喂给机械臂。
– 时空双维控制：静态手势管“夹爪开合”，动态手势管“任务启停”，互不干扰，交互丝滑。
🏗️ 系统架构
系统采用边缘侧感知 + 中心侧决策的分布式架构，不仅缓解了边缘设备的算力焦虑，还保证了控制的实时性。
| 节点 | 硬件载体 | 核心职责 | 技术栈 |
|---|---|---|---|
| 👁️ 感知层 | Jetson Orin Nano + Gemini2 | 手部区域检测、3D坐标提取、静态手势识别 | PyOrbbecSDK, MediaPipe, OpenCV |
| 🧠 决策层 | PC 电脑 | 动态手势识别（时序动作）、坐标变换 | TensorFlow/Keras, ResNet3D |
| 🦾 执行层 | DOBOT Magician | 定点移动、夹爪开合 | DOBOT SDK, Python |
🔄 数据流转：
🧩 核心模块说明
1. 🎯 3D 坐标提取模块（视觉核心）
不玩虚的，直接上硬核深度对齐！系统并非简单依赖 2D 画面，而是通过 Orbbec SDK 获取对齐的深度帧。
– 流程：
def main():
pipeline = Pipeline()
device = pipeline.get_device()
device_info = device.get_device_info()
device_pid = device_info.get_pid()
config = Config()
parser = argparse.ArgumentParser()
parser.add_argument("-m", "--mode",
help="align mode, HW=hardware mode,SW=software mode,NONE=disable align",
type=str, default='HW')
parser.add_argument("-s", "--enable_sync", help="enable sync", type=bool, default=True)
args = parser.parse_args()
align_mode = args.mode
enable_sync = args.enable_sync
try:
profile_list = pipeline.get_stream_profile_list(OBSensorType.COLOR_SENSOR)
color_profile = profile_list.get_default_video_stream_profile()
config.enable_stream(color_profile)
profile_list = pipeline.get_stream_profile_list(OBSensorType.DEPTH_SENSOR)
assert profile_list is not None
depth_profile = profile_list.get_default_video_stream_profile()
assert depth_profile is not None
print("color profile : {}x{}@{}_{}".format(color_profile.get_width(),
color_profile.get_height(),
color_profile.get_fps(),
color_profile.get_format()))
print("depth profile : {}x{}@{}_{}".format(depth_profile.get_width(),
depth_profile.get_height(),
depth_profile.get_fps(),
depth_profile.get_format()))
config.enable_stream(depth_profile)
except Exception as e:
print(e)
return
if align_mode == 'HW':
if device_pid == 0x066B:
# Femto Mega does not support hardware D2C, and it is changed to software D2C
config.set_align_mode(OBAlignMode.SW_MODE)
else:
config.set_align_mode(OBAlignMode.HW_MODE)
elif align_mode == 'SW':
config.set_align_mode(OBAlignMode.SW_MODE)
else:
config.set_align_mode(OBAlignMode.DISABLE)
if enable_sync:
try:
pipeline.enable_frame_sync()
except Exception as e:
print(e)
try:
pipeline.start(config) # 启动相机流
# pipeline.start(config)
except Exception as e:
print(e)
return
# 初始化mediapipe手部检测模型
mp_hands = mp.solutions.hands.Hands(static_image_mode=True, max_num_hands=1, min_detection_confidence=0.5,
min_tracking_confidence=0.5)
# 创建并启动图像处理线程
process_thread = Thread(target=process_frames, args=(pipeline, mp_hands))
process_thread.start()
# 等待图像处理线程结束
process_thread.join()
# 停止相机流和关闭窗口
pipeline.stop()
cv2.destroyAllWindows()– 技术取舍：相比于纯点云分割，2D关键点+深度查询的方式在 Jetson 这种嵌入式板卡上帧率更高、延迟更低。
2. ✋ 静态手势识别（状态控制）
基于 MediaPipe 提取的手部 21 个关键点，通过判断指尖与掌心的距离关系，快速判定握拳（抓取）或张手（释放）。
def detect_and_display_hands(image, mp_hands):
# 处理RGB图像并检测手部
global center_x
global center_y
results = mp_hands.process(image)
fist_detected = False # Initialize the fist detection flag
if results.multi_hand_landmarks:
for hand_landmarks in results.multi_hand_landmarks:
hand_points = []
for landmark in hand_landmarks.landmark:
# 计算手部关键点的像素坐标
cx, cy = int(landmark.x * image.shape[1]), int(landmark.y * image.shape[0])
hand_points.append((cx, cy))
# 在关键点位置画圆点
cv2.circle(image, (cx, cy), 5, (255, 0, 0), cv2.FILLED)
# 获取手部边界框
x_min, x_max, y_min, y_max = float('inf'), -float('inf'), float('inf'), -float('inf')
for x, y in hand_points:
x_min = min(x_min, x)
x_max = max(x_max, x)
y_min = min(y_min, y)
y_max = max(y_max, y)
# 在图像上画出手部边界框
cv2.rectangle(image, (x_min, y_min), (x_max, y_max), (0, 255, 0), 2)
# 计算手部中心的坐标
center_x, center_y = (x_min + x_max) // 2, (y_min + y_max) // 2
# 绘制手部关键点之间的连线
connections = [
(0, 1), (1, 2), (2, 3), (3, 4), # 手指 1
(0, 5), (5, 6), (6, 7), (7, 8), # 手指 2
(0, 9), (9, 10), (10, 11), (11, 12), # 手指 3
(0, 13), (13, 14), (14, 15), (15, 16), # 手指 4
(0, 17), (17, 18), (18, 19), (19, 20), # 手掌
]
for connection in connections:
point1_id, point2_id = connection
point1 = hand_points[point1_id]
point2 = hand_points[point2_id]
# 在图像上绘制连接线
cv2.line(image, point1, point2, (0, 255, 255), 2)
# Check for fist gesture
if len(hand_points) == 21: # Check if all hand landmarks are detected
fingers_up = [0, 0, 0, 0, 0] # Initialize finger status
# Thumb
if hand_points[4][1] < hand_points[3][1]:
fingers_up[0] = 1
# Other fingers
for i in range(1, 5):
if hand_points[i * 4][1] < hand_points[i * 4 - 2][1]:
fingers_up[i] = 1
if sum(fingers_up) == 0: # If all fingers are down, it's a fist gesture
fist_detected = True– 优势：无需经过重型网络，毫秒级响应，完美契合夹爪控制这种需要高频状态确认的场景。
3. 🎬 动态手势识别（时序决策）
静态手势管当下，动态手势管意图。系统采用 3D ResNet 提取视频帧的时空特征。
– 模型架构：支持标准 CNN3D 与 Resnet3DBuilder。通过 DataLoader 将视频帧序列 (nb_frames, H, W, 3) 喂入网络，捕捉手部在一段时间内的动作轨迹（如挥手、比心等）。
– 工程细节：使用了自定义的 ImageDataGenerator 进行数据增强，并支持多线程加载，保证训练时 GPU 不闲置。
class ImageDataGenerator(object):
"""Generate batches of tensor image data with real-time data augmentation.
The data will be looped over (in batches).
# Arguments
featurewise_center: Boolean.
Set input mean to 0 over the dataset, feature-wise.
samplewise_center: Boolean. Set each sample mean to 0.
featurewise_std_normalization: Boolean.
Divide inputs by std of the dataset, feature-wise.
samplewise_std_normalization: Boolean. Divide each input by its std.
zca_epsilon: epsilon for ZCA whitening. Default is 1e-6.
zca_whitening: Boolean. Apply ZCA whitening.
rotation_range: Int. Degree range for random rotations.
width_shift_range: Float, 1-D array-like or int
- float: fraction of total width, if < 1, or pixels if >= 1.
- 1-D array-like: random elements from the array.
- int: integer number of pixels from interval
`(-width_shift_range, +width_shift_range)`
- With `width_shift_range=2` possible values
are integers `[-1, 0, +1]`,
same as with `width_shift_range=[-1, 0, +1]`,
while with `width_shift_range=1.0` possible values are floats
in the interval [-1.0, +1.0).
height_shift_range: Float, 1-D array-like or int
- float: fraction of total height, if < 1, or pixels if >= 1.
- 1-D array-like: random elements from the array.
- int: integer number of pixels from interval
`(-height_shift_range, +height_shift_range)`
- With `height_shift_range=2` possible values
are integers `[-1, 0, +1]`,
same as with `height_shift_range=[-1, 0, +1]`,
while with `height_shift_range=1.0` possible values are floats
in the interval [-1.0, +1.0).
brightness_range: Tuple or list of two floats. Range for picking
a brightness shift value from.
shear_range: Float. Shear Intensity
(Shear angle in counter-clockwise direction in degrees)
zoom_range: Float or [lower, upper]. Range for random zoom.
If a float, `[lower, upper] = [1-zoom_range, 1+zoom_range]`.
channel_shift_range: Float. Range for random channel shifts.
fill_mode: One of {"constant", "nearest", "reflect" or "wrap"}.
Default is 'nearest'.
Points outside the boundaries of the input are filled
according to the given mode:
- 'constant': kkkkkkkk|abcd|kkkkkkkk (cval=k)
- 'nearest': aaaaaaaa|abcd|dddddddd
- 'reflect': abcddcba|abcd|dcbaabcd
- 'wrap': abcdabcd|abcd|abcdabcd
cval: Float or Int.
Value used for points outside the boundaries
when `fill_mode = "constant"`.
horizontal_flip: Boolean. Randomly flip inputs horizontally.
vertical_flip: Boolean. Randomly flip inputs vertically.
rescale: rescaling factor. Defaults to None.
If None or 0, no rescaling is applied,
otherwise we multiply the data by the value provided
(after applying all other transformations).
preprocessing_function: function that will be implied on each input.
The function will run after the image is resized and augmented.
The function should take one argument:
one image (Numpy tensor with rank 3),
and should output a Numpy tensor with the same shape.
data_format: Image data format,
either "channels_first" or "channels_last".
"channels_last" mode means that the images should have shape
`(samples, height, width, channels)`,
"channels_first" mode means that the images should have shape
`(samples, channels, height, width)`.
It defaults to the `image_data_format` value found in your
Keras config file at `~/.keras/keras.json`.
If you never set it, then it will be "channels_last".
validation_split: Float. Fraction of images reserved for validation
(strictly between 0 and 1).
dtype: Dtype to use for the generated arrays.
# Examples
Example of using `.flow(x, y)`:
```python
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
y_train = np_utils.to_categorical(y_train, num_classes)
y_test = np_utils.to_categorical(y_test, num_classes)
datagen = ImageDataGenerator(
featurewise_center=True,
featurewise_std_normalization=True,
rotation_range=20,
width_shift_range=0.2,
height_shift_range=0.2,
horizontal_flip=True)
# compute quantities required for featurewise normalization
# (std, mean, and principal components if ZCA whitening is applied)
datagen.fit(x_train)
# fits the model on batches with real-time data augmentation:
model.fit_generator(datagen.flow(x_train, y_train, batch_size=32),
steps_per_epoch=len(x_train) / 32, epochs=epochs)
# here's a more "manual" example
for e in range(epochs):
print('Epoch', e)
batches = 0
for x_batch, y_batch in datagen.flow(x_train, y_train, batch_size=32):
model.fit(x_batch, y_batch)
batches += 1
if batches >= len(x_train) / 32:
# we need to break the loop by hand because
# the generator loops indefinitely
breakExample of using .flow_from_directory(directory):
train_datagen = ImageDataGenerator(
rescale=1./255,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True)
test_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_directory(
'data/train',
target_size=(150, 150),
batch_size=32,
class_mode='binary')
validation_generator = test_datagen.flow_from_directory(
'data/validation',
target_size=(150, 150),
batch_size=32,
class_mode='binary')
model.fit_generator(
train_generator,
steps_per_epoch=2000,
epochs=50,
validation_data=validation_generator,
validation_steps=800)Example of transforming images and masks together.
# we create two instances with the same arguments
data_gen_args = dict(featurewise_center=True,
featurewise_std_normalization=True,
rotation_range=90,
width_shift_range=0.1,
height_shift_range=0.1,
zoom_range=0.2)
image_datagen = ImageDataGenerator(**data_gen_args)
mask_datagen = ImageDataGenerator(**data_gen_args)
# Provide the same seed and keyword arguments to the fit and flow methods
seed = 1
image_datagen.fit(images, augment=True, seed=seed)
mask_datagen.fit(masks, augment=True, seed=seed)
image_generator = image_datagen.flow_from_directory(
'data/images',
class_mode=None,
seed=seed)
mask_generator = mask_datagen.flow_from_directory(
'data/masks',
class_mode=None,
seed=seed)
# combine generators into one which yields image and masks
train_generator = zip(image_generator, mask_generator)
model.fit_generator(
train_generator,
steps_per_epoch=2000,
epochs=50)Example of using “.flow_from_dataframe(dataframe, directory, x_col, y_col, has_ext)“:
train_df = pandas.read_csv("./train.csv")
valid_df = pandas.read_csv("./valid.csv")
train_datagen = ImageDataGenerator(
rescale=1./255,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True)
test_datagen = ImageDataGenerator(rescale=1./255)
train_generator = train_datagen.flow_from_dataframe(
dataframe=train_df,
directory='data/train',
x_col="filename",
y_col="class",
has_ext=True,
target_size=(150, 150),
batch_size=32,
class_mode='binary')
validation_generator = test_datagen.flow_from_dataframe(
dataframe=valid_df,
directory='data/validation',
x_col="filename",
y_col="class",
has_ext=True,
target_size=(150, 150),
batch_size=32,
class_mode='binary')
model.fit_generator(
train_generator,
steps_per_epoch=2000,
epochs=50,
validation_data=validation_generator,
validation_steps=800)"""
def __init__(self, featurewise_center=False, samplewise_center=False, featurewise_std_normalization=False, samplewise_std_normalization=False, zca_whitening=False, zca_epsilon=1e-6, rotation_range=0, width_shift_range=0., height_shift_range=0., brightness_range=None, shear_range=0., zoom_range=0., channel_shift_range=0., fill_mode='nearest', cval=0., horizontal_flip=False, vertical_flip=False, rescale=None, preprocessing_function=None, data_format='channels_last', validation_split=0.0, dtype='float32'): self.featurewise_center = featurewise_center self.samplewise_center = samplewise_center self.featurewise_std_normalization = featurewise_std_normalization self.samplewise_std_normalization = samplewise_std_normalization self.zca_whitening = zca_whitening self.zca_epsilon = zca_epsilon self.rotation_range = rotation_range self.width_shift_range = width_shift_range self.height_shift_range = height_shift_range self.brightness_range = brightness_range self.shear_range = shear_range self.zoom_range = zoom_range self.channel_shift_range = channel_shift_range self.fill_mode = fill_mode self.cval = cval self.horizontal_flip = horizontal_flip self.vertical_flip = vertical_flip self.rescale = rescale self.preprocessing_function = preprocessing_function self.dtype = dtype
if data_format not in {'channels_last', 'channels_first'}: raise ValueError( 'data_format should be "channels_last" ' '(channel after row and column) or ' '"channels_first" (channel before row and column). ' 'Received: %s' % data_format) self.data_format = data_format if data_format == 'channels_first': self.channel_axis = 1 self.row_axis = 2 self.col_axis = 3 if data_format == 'channels_last': self.channel_axis = 3 self.row_axis = 1 self.col_axis = 2 if validation_split and not 0 < validation_split < 1: raise ValueError( 'validation_split must be strictly between 0 and 1. ' ' Received: %s' % validation_split) self._validation_split = validation_split
self.mean = None self.std = None self.principal_components = None
if np.isscalar(zoom_range): self.zoom_range = [1 – zoom_range, 1 + zoom_range] elif len(zoom_range) == 2: self.zoom_range = [zoom_range[0], zoom_range[1]] else: raise ValueError('zoom_range should be a float or ' 'a tuple or list of two floats. ' 'Received: %s' % (zoom_range,)) if zca_whitening: if not featurewise_center: self.featurewise_center = True warnings.warn('This ImageDataGenerator specifies ' 'zca_whitening, which overrides ' 'setting of featurewise_center.') if featurewise_std_normalization: self.featurewise_std_normalization = False warnings.warn('This ImageDataGenerator specifies ' 'zca_whitening ' 'which overrides setting of' 'featurewise_std_normalization.') if featurewise_std_normalization: if not featurewise_center: self.featurewise_center = True warnings.warn('This ImageDataGenerator specifies ' 'featurewise_std_normalization, ' 'which overrides setting of ' 'featurewise_center.') if samplewise_std_normalization: if not samplewise_center: self.samplewise_center = True warnings.warn('This ImageDataGenerator specifies ' 'samplewise_std_normalization, ' 'which overrides setting of ' 'samplewise_center.')
def flow(self, x, y=None, batch_size=32, shuffle=True, sample_weight=None, seed=None, save_to_dir=None, save_prefix='', save_format='png', subset=None): """Takes data & label arrays, generates batches of augmented data.
# Arguments x: Input data. Numpy array of rank 4 or a tuple. If tuple, the first element should contain the images and the second element another numpy array or a list of numpy arrays that gets passed to the output without any modifications. Can be used to feed the model miscellaneous data along with the images. In case of grayscale data, the channels axis of the image array should have value 1, in case of RGB data, it should have value 3, and in case of RGBA data, it should have value 4. y: Labels. batch_size: Int (default: 32). shuffle: Boolean (default: True). sample_weight: Sample weights. seed: Int (default: None). save_to_dir: None or str (default: None). This allows you to optionally specify a directory to which to save the augmented pictures being generated (useful for visualizing what you are doing). save_prefix: Str (default: ''). Prefix to use for filenames of saved pictures (only relevant if save_to_dir is set). save_format: one of "png", "jpeg" (only relevant if save_to_dir is set). Default: "png". subset: Subset of data ("training" or "validation") if validation_split is set in ImageDataGenerator.
# Returns An Iterator yielding tuples of (x, y) where x is a numpy array of image data (in the case of a single image input) or a list of numpy arrays (in the case with additional inputs) and y is a numpy array of corresponding labels. If 'sample_weight' is not None, the yielded tuples are of the form (x, y, sample_weight). If y is None, only the numpy array x is returned. """ return NumpyArrayIterator( x, y, self, batch_size=batch_size, shuffle=shuffle, sample_weight=sample_weight, seed=seed, data_format=self.data_format, save_to_dir=save_to_dir, save_prefix=save_prefix, save_format=save_format, subset=subset)
def flow_from_directory(self, directory, target_size=(256, 256), color_mode='rgb', classes=None, class_mode='categorical', batch_size=32, shuffle=True, seed=None, save_to_dir=None, save_prefix='', save_format='png', follow_links=False, subset=None, interpolation='nearest'): """Takes the path to a directory & generates batches of augmented data.
# Arguments directory: Path to the target directory. It should contain one subdirectory per class. Any PNG, JPG, BMP, PPM or TIF images inside each of the subdirectories directory tree will be included in the generator. See this script for more details. target_size: Tuple of integers (height, width), default: (256, 256). The dimensions to which all images found will be resized. color_mode: One of "grayscale", "rbg", "rgba". Default: "rgb". Whether the images will be converted to have 1, 3, or 4 channels. classes: Optional list of class subdirectories (e.g. ['dogs', 'cats']). Default: None. If not provided, the list of classes will be automatically inferred from the subdirectory names/structure under directory, where each subdirectory will be treated as a different class (and the order of the classes, which will map to the label indices, will be alphanumeric). The dictionary containing the mapping from class names to class indices can be obtained via the attribute class_indices. class_mode: One of "categorical", "binary", "sparse", "input", or None. Default: "categorical". Determines the type of label arrays that are returned:
- "categorical" will be 2D one-hot encoded labels,
- "binary" will be 1D binary labels,
"sparse" will be 1D integer labels,
- "input" will be images identical
to input images (mainly used to work with autoencoders).
- If None, no labels are returned
(the generator will only yield batches of image data, which is useful to use with model.predict_generator(), model.evaluate_generator(), etc.). Please note that in case of class_mode None, the data still needs to reside in a subdirectory of directory for it to work correctly. batch_size: Size of the batches of data (default: 32). shuffle: Whether to shuffle the data (default: True) seed: Optional random seed for shuffling and transformations. save_to_dir: None or str (default: None). This allows you to optionally specify a directory to which to save the augmented pictures being generated (useful for visualizing what you are doing). save_prefix: Str. Prefix to use for filenames of saved pictures (only relevant if save_to_dir is set). save_format: One of "png", "jpeg" (only relevant if save_to_dir is set). Default: "png". follow_links: Whether to follow symlinks inside class subdirectories (default: False). subset: Subset of data ("training" or "validation") if validation_split is set in ImageDataGenerator. interpolation: Interpolation method used to resample the image if the target size is different from that of the loaded image. Supported methods are "nearest", "bilinear", and "bicubic". If PIL version 1.1.3 or newer is installed, "lanczos" is also supported. If PIL version 3.4.0 or newer is installed, "box" and "hamming" are also supported. By default, "nearest" is used.
# Returns A DirectoryIterator yielding tuples of (x, y) where x is a numpy array containing a batch of images with shape (batch_size, *target_size, channels) and y is a numpy array of corresponding labels. """ return DirectoryIterator( directory, self, target_size=target_size, color_mode=color_mode, classes=classes, class_mode=class_mode, data_format=self.data_format, batch_size=batch_size, shuffle=shuffle, seed=seed, save_to_dir=save_to_dir, save_prefix=save_prefix, save_format=save_format, follow_links=follow_links, subset=subset, interpolation=interpolation)
def flow_from_dataframe(self, dataframe, directory, x_col="filename", y_col="class", has_ext=True, target_size=(256, 256), color_mode='rgb', classes=None, class_mode='categorical', batch_size=32, shuffle=True, seed=None, save_to_dir=None, save_prefix='', save_format='png', subset=None, interpolation='nearest'): """Takes the dataframe and the path to a directory and generates batches of augmented/normalized data.
# A simple tutorial can be found at: http://bit.ly/keras_flow_from_dataframe
# Arguments dataframe: Pandas dataframe containing the filenames of the images in a column and classes in another or column/s that can be fed as raw target data. directory: string, path to the target directory that contains all the images mapped in the dataframe. x_col: string, column in the dataframe that contains the filenames of the target images. y_col: string or list of strings,columns in the dataframe that will be the target data. has_ext: bool, True if filenames in dataframe[x_col] has filename extensions,else False. target_size: tuple of integers (height, width), default: (256, 256). The dimensions to which all images found will be resized. color_mode: one of "grayscale", "rbg". Default: "rgb". Whether the images will be converted to have 1 or 3 color channels. classes: optional list of classes (e.g. ['dogs', 'cats']). Default: None. If not provided, the list of classes will be automatically inferred from the y_col, which will map to the label indices, will be alphanumeric). The dictionary containing the mapping from class names to class indices can be obtained via the attribute class_indices. class_mode: one of "categorical", "binary", "sparse", "input", "other" or None. Default: "categorical". Determines the type of label arrays that are returned:
"categorical"will be 2D one-hot encoded labels,"binary"will be 1D binary labels,"sparse"will be 1D integer labels,"input"will be images identical
to input images (mainly used to work with autoencoders).
"other"will be numpy array of y_col data- None, no labels are returned (the generator will only
yield batches of image data, which is useful to use model.predict_generator(), model.evaluate_generator(), etc.). batch_size: size of the batches of data (default: 32). shuffle: whether to shuffle the data (default: True) seed: optional random seed for shuffling and transformations. save_to_dir: None or str (default: None). This allows you to optionally specify a directory to which to save the augmented pictures being generated (useful for visualizing what you are doing). save_prefix: str. Prefix to use for filenames of saved pictures (only relevant if save_to_dir is set). save_format: one of "png", "jpeg" (only relevant if save_to_dir is set). Default: "png". follow_links: whether to follow symlinks inside class subdirectories (default: False). subset: Subset of data ("training" or "validation") if validation_split is set in ImageDataGenerator. interpolation: Interpolation method used to resample the image if the target size is different from that of the loaded image. Supported methods are "nearest", "bilinear", and "bicubic". If PIL version 1.1.3 or newer is installed, "lanczos" is also supported. If PIL version 3.4.0 or newer is installed, "box" and "hamming" are also supported. By default, "nearest" is used.
# Returns A DataFrameIterator yielding tuples of (x, y) where x is a numpy array containing a batch of images with shape (batch_size, *target_size, channels) and y is a numpy array of corresponding labels. """
return DataFrameIterator(dataframe, directory, self, x_col=x_col, y_col=y_col, has_ext=has_ext, target_size=target_size, color_mode=color_mode, classes=classes, class_mode=class_mode, data_format=self.data_format, batch_size=batch_size, shuffle=shuffle, seed=seed, save_to_dir=save_to_dir, save_prefix=save_prefix, save_format=save_format, subset=subset, interpolation=interpolation)
def flow_video_from_dataframe(self, dataframe, directory, x_col="filename", y_col="class", has_ext=True, target_size=(256, 256), color_mode='rgb', classes=None, path_classes=None, class_mode='categorical', batch_size=32, nb_frames=1, shuffle=True, seed=None, skip=0, save_to_dir=None, save_prefix='', save_format='png', subset=None, interpolation='nearest'): """Takes the dataframe and the path to a directory and generates batches of augmented/normalized data.
# A simple tutorial can be found at: http://bit.ly/keras_flow_from_dataframe
# Arguments dataframe: Pandas dataframe containing the filenames of the images in a column and classes in another or column/s that can be fed as raw target data. directory: string, path to the target directory that contains all the images mapped in the dataframe. x_col: string, column in the dataframe that contains the filenames of the target images. y_col: string or list of strings,columns in the dataframe that will be the target data. has_ext: bool, True if filenames in dataframe[x_col] has filename extensions,else False. target_size: tuple of integers (height, width), default: (256, 256). The dimensions to which all images found will be resized. color_mode: one of "grayscale", "rbg". Default: "rgb". Whether the images will be converted to have 1 or 3 color channels. classes: optional list of classes (e.g. ['dogs', 'cats']). Default: None. If not provided, the list of classes will be automatically inferred from the y_col, which will map to the label indices, will be alphanumeric). The dictionary containing the mapping from class names to class indices can be obtained via the attribute class_indices. class_mode: one of "categorical", "binary", "sparse", "input", "other" or None. Default: "categorical". Determines the type of label arrays that are returned:
"categorical"will be 2D one-hot encoded labels,"binary"will be 1D binary labels,"sparse"will be 1D integer labels,"input"will be images identical
to input images (mainly used to work with autoencoders).
"other"will be numpy array of y_col data- None, no labels are returned (the generator will only
yield batches of image data, which is useful to use model.predict_generator(), model.evaluate_generator(), etc.). batch_size: size of the batches of data (default: 32). nb_frames: Integer, number of frames to use for each video. shuffle: whether to shuffle the data (default: True) seed: optional random seed for shuffling and transformations. skip: Integer, number of frames to skip between each frame of the data video folder save_to_dir: None or str (default: None). This allows you to optionally specify a directory to which to save the augmented pictures being generated (useful for visualizing what you are doing). save_prefix: str. Prefix to use for filenames of saved pictures (only relevant if save_to_dir is set). save_format: one of "png", "jpeg" (only relevant if save_to_dir is set). Default: "png". follow_links: whether to follow symlinks inside class subdirectories (default: False). subset: Subset of data ("training" or "validation") if validation_split is set in ImageDataGenerator. interpolation: Interpolation method used to resample the image if the target size is different from that of the loaded image. Supported methods are "nearest", "bilinear", and "bicubic". If PIL version 1.1.3 or newer is installed, "lanczos" is also supported. If PIL version 3.4.0 or newer is installed, "box" and "hamming" are also supported. By default, "nearest" is used.
# Returns A DataFrameIterator yielding tuples of (x, y) where x is a numpy array containing a batch of images with shape (batch_size, *target_size, channels) and y is a numpy array of corresponding labels. """ return DataFrameIterator(dataframe, directory, self, x_col=x_col, y_col=y_col, has_ext=has_ext, target_size=target_size, color_mode=color_mode, classes=classes, path_classes=path_classes, class_mode=class_mode, data_format=self.data_format, batch_size=batch_size, nb_frames=nb_frames, shuffle=shuffle, seed=seed, skip=skip, save_to_dir=save_to_dir, save_prefix=save_prefix, save_format=save_format, subset=subset, interpolation=interpolation)
def standardize(self, x): """Applies the normalization configuration to a batch of inputs.
# Arguments x: Batch of inputs to be normalized.
# Returns The inputs, normalized. """ if self.preprocessing_function: x = self.preprocessing_function(x) if self.rescale: x *= self.rescale if self.samplewise_center: x -= np.mean(x, keepdims=True) if self.samplewise_std_normalization: x /= (np.std(x, keepdims=True) + 1e-6)
if self.featurewise_center: if self.mean is not None: x -= self.mean else: warnings.warn('This ImageDataGenerator specifies ' 'featurewise_center, but it hasn't ' 'been fit on any training data. Fit it ' 'first by calling .fit(numpy_data).') if self.featurewise_std_normalization: if self.std is not None: x /= (self.std + 1e-6) else: warnings.warn('This ImageDataGenerator specifies ' 'featurewise_std_normalization, ' 'but it hasn't ' 'been fit on any training data. Fit it ' 'first by calling .fit(numpy_data).') if self.zca_whitening: if self.principal_components is not None: flatx = np.reshape(x, (-1, np.prod(x.shape[-3:]))) whitex = np.dot(flatx, self.principal_components) x = np.reshape(whitex, x.shape) else: warnings.warn('This ImageDataGenerator specifies ' 'zca_whitening, but it hasn't ' 'been fit on any training data. Fit it ' 'first by calling .fit(numpy_data).') return x
def get_random_transform(self, img_shape, seed=None): """Generates random parameters for a transformation.
# Arguments seed: Random seed. img_shape: Tuple of integers. Shape of the image that is transformed.
# Returns A dictionary containing randomly chosen parameters describing the transformation. """ img_row_axis = self.row_axis – 1 img_col_axis = self.col_axis – 1
if seed is not None: np.random.seed(seed)
if self.rotation_range: theta = np.random.uniform( -self.rotation_range, self.rotation_range) else: theta = 0
if self.height_shift_range: try: # 1-D array-like or int tx = np.random.choice(self.height_shift_range) tx = np.random.choice([-1, 1]) except ValueError: # floating point tx = np.random.uniform(-self.height_shift_range, self.height_shift_range) if np.max(self.height_shift_range) < 1: tx = img_shape[img_row_axis] else: tx = 0
if self.width_shift_range: try: # 1-D array-like or int ty = np.random.choice(self.width_shift_range) ty = np.random.choice([-1, 1]) except ValueError: # floating point ty = np.random.uniform(-self.width_shift_range, self.width_shift_range) if np.max(self.width_shift_range) < 1: ty = img_shape[img_col_axis] else: ty = 0
if self.shear_range: shear = np.random.uniform( -self.shear_range, self.shear_range) else: shear = 0
if self.zoom_range[0] == 1 and self.zoom_range[1] == 1: zx, zy = 1, 1 else: zx, zy = np.random.uniform( self.zoom_range[0], self.zoom_range[1], 2)
flip_horizontal = (np.random.random() < 0.5) self.horizontal_flip flip_vertical = (np.random.random() < 0.5) self.vertical_flip
channel_shift_intensity = None if self.channel_shift_range != 0: channel_shift_intensity = np.random.uniform(-self.channel_shift_range, self.channel_shift_range)
brightness = None if self.brightness_range is not None: if len(self.brightness_range) != 2: raise ValueError( '`brightness_range should be tuple or list of two floats. ' 'Received: %s' % (self.brightness_range,)) brightness = np.random.uniform(self.brightness_range[0], self.brightness_range[1])
transform_parameters = {'theta': theta, 'tx': tx, 'ty': ty, 'shear': shear, 'zx': zx, 'zy': zy, 'flip_horizontal': flip_horizontal, 'flip_vertical': flip_vertical, 'channel_shift_intensity': channel_shift_intensity, 'brightness': brightness}
return transform_parameters
def apply_transform(self, x, transform_parameters): """Applies a transformation to an image according to given parameters.
# Arguments x: 3D tensor, single image. transform_parameters: Dictionary with string – parameter pairs describing the transformation. Currently, the following parameters from the dictionary are used:
'theta': Float. Rotation angle in degrees.'tx': Float. Shift in the x direction.'ty': Float. Shift in the y direction.'shear': Float. Shear angle in degrees.'zx': Float. Zoom in the x direction.'zy': Float. Zoom in the y direction.'flip_horizontal': Boolean. Horizontal flip.'flip_vertical': Boolean. Vertical flip.'channel_shift_intencity': Float. Channel shift intensity.'brightness': Float. Brightness shift intensity.
# Returns A transformed version of the input (same shape). """ # x is a single image, so it doesn't have image number at index 0 img_row_axis = self.row_axis – 1 img_col_axis = self.col_axis – 1 img_channel_axis = self.channel_axis – 1
x = apply_affine_transform(x, transform_parameters.get('theta', 0), transform_parameters.get('tx', 0), transform_parameters.get('ty', 0), transform_parameters.get('shear', 0), transform_parameters.get('zx', 1), transform_parameters.get('zy', 1), row_axis=img_row_axis, col_axis=img_col_axis, channel_axis=img_channel_axis, fill_mode=self.fill_mode, cval=self.cval)
if transform_parameters.get('channel_shift_intensity') is not None: x = apply_channel_shift(x, transform_parameters['channel_shift_intensity'], img_channel_axis)
if transform_parameters.get('flip_horizontal', False): x = flip_axis(x, img_col_axis)
if transform_parameters.get('flip_vertical', False): x = flip_axis(x, img_row_axis)
if transform_parameters.get('brightness') is not None: x = apply_brightness_shift(x, transform_parameters['brightness'])
return x
def random_transform(self, x, seed=None): """Applies a random transformation to an image.
# Arguments x: 3D tensor, single image. seed: Random seed.
# Returns A randomly transformed version of the input (same shape). """ params = self.get_random_transform(x.shape, seed) return self.apply_transform(x, params)
def fit(self, x, augment=False, rounds=1, seed=None): """Fits the data generator to some sample data.
This computes the internal data stats related to the data-dependent transformations, based on an array of sample data.
Only required if featurewise_center or featurewise_std_normalization or zca_whitening are set to True.
# Arguments x: Sample data. Should have rank 4. In case of grayscale data, the channels axis should have value 1, in case of RGB data, it should have value 3, and in case of RGBA data, it should have value 4. augment: Boolean (default: False). Whether to fit on randomly augmented samples. rounds: Int (default: 1). If using data augmentation (augment=True), this is how many augmentation passes over the data to use. seed: Int (default: None). Random seed. """ x = np.asarray(x, dtype=self.dtype) if x.ndim != 4: raise ValueError('Input to .fit() should have rank 4. ' 'Got array with shape: ' + str(x.shape)) if x.shape[self.channel_axis] not in {1, 3, 4}: warnings.warn( 'Expected input to be images (as Numpy array) ' 'following the data format convention "' + self.data_format + '" (channels on axis ' + str(self.channel_axis) + '), i.e. expected ' 'either 1, 3 or 4 channels on axis ' + str(self.channel_axis) + '. ' 'However, it was passed an array with shape ' + str(x.shape) + ' (' + str(x.shape[self.channel_axis]) + ' channels).')
if seed is not None: np.random.seed(seed)
x = np.copy(x) if augment: ax = np.zeros( tuple([rounds x.shape[0]] + list(x.shape)[1:]), dtype=self.dtype) for r in range(rounds): for i in range(x.shape[0]): ax[i + r x.shape[0]] = self.random_transform(x[i]) x = ax
if self.featurewise_center: self.mean = np.mean(x, axis=(0, self.row_axis, self.col_axis)) broadcast_shape = [1, 1, 1] broadcast_shape[self.channel_axis – 1] = x.shape[self.channel_axis] self.mean = np.reshape(self.mean, broadcast_shape) x -= self.mean
if self.featurewise_std_normalization: self.std = np.std(x, axis=(0, self.row_axis, self.col_axis)) broadcast_shape = [1, 1, 1] broadcast_shape[self.channel_axis – 1] = x.shape[self.channel_axis] self.std = np.reshape(self.std, broadcast_shape) x /= (self.std + 1e-6)
if self.zca_whitening: if scipy is None: raise ImportError('Using zca_whitening requires SciPy. ' 'Install SciPy.') flat_x = np.reshape( x, (x.shape[0], x.shape[1] x.shape[2] x.shape[3])) sigma = np.dot(flat_x.T, flat_x) / flat_x.shape[0] u, s, _ = scipy.linalg.svd(sigma) s_inv = 1. / np.sqrt(s[np.newaxis] + self.zca_epsilon) self.principal_components = (u * s_inv).dot(u.T)
## **📑 设计要点**
整个交互逻辑设计得非常符合人类直觉，形成了**“跟随-抓取-移动-释放”**的完整闭环：
1. **👀 锁定目标**：相机视野内出现手部，系统唤醒，机械臂末端实时跟随手部 3D 坐标移动。
2. **✊ 执行抓取**：手部移动至待抓取物上方，用户握拳，系统识别为“抓取”指令，夹爪闭合。
3. **🚚 物料搬运**：用户保持握拳，手部移动，机械臂夹持物体跟随移动至目标区域。
4. **🖐️ 释放物体**：用户张开手掌，系统识别为“释放”指令，夹爪松开，任务完成。
### **🛠️ 硬件选型巧思**
- **Orbbec** **Gemini2**：搭载 MX6600 芯片，0.15-10m 零盲区测量，完美解决机械臂近场操作时的深度盲区问题。
- **Jetson Orin Nano**：40 TOPS 的 AI 算力，跑 MediaPipe 和图像预处理绰绰有余，功耗仅 5-15W，适合桌面级机械臂场景。
## **🚀 使用示例**
项目代码完全使用 Python 编写，基于配置文件驱动，对二次开发非常友好。
### **Step 1: 启动边缘侧感知（Jetson Orin Nano）**
运行相机节点，获取手部坐标与静态手势：
```Bash
python test-budongle.py
# 终端将实时输出：
# 手部中心坐标： x, y
# 深度距离: zStep 2: 训练自定义动态手势（PC 端）
修改 config.ini 中的路径，放入你的手势视频数据集，即可训练自己的 3D ResNet：
python train.py --config config.ini
# 模型会自动保存为 HDF5 或 JSON+Weights 格式Step 3: 启动动态识别与控制（PC 端）
加载预训练模型，开启摄像头与机械臂的联动：
python myhanuman.py
# 系统将加载 CNN 模型，实时捕捉动态手势并映射为控制指令🧭 后续规划
虽然系统已经跑通了闭环，但在具身智能的坑里，永远有优化的空间：
– ⏱️ 端到端延迟优化：目前 PC 与 Jetson 间存在通信开销，计划将动态手势模型量化（TensorRT）后全量部署至 Jetson，实现纯边缘闭环。
– 🛡️ 抗遮挡与手部追踪：当手部被物体遮挡时关键点容易抖动，后续考虑引入卡尔曼滤波或 ByteTrack 等追踪算法平滑坐标。
– 🤖 多臂协同：当前仅支持单臂，未来探索双手交互识别，控制双臂进行更复杂的装配任务。
> 💡 结语：从 2D 像素到 3D 空间，从静态握拳到动态挥舞，这个项目展示了视觉感知如何赋予机器“眼”与“手”的协调能力。欢迎 Star & Fork，一起来教机器人做手语！