对象检测¶
对象检测服务支持使用 Multimedia API 从图像中提取车辆、人员、图案、子图像等对象。
默认情况下,ecu.test 支持用于识别图像内子图像的简单算法(模板匹配和基于特征的匹配)。也可以使用 UserPyModules 集成额外的对象检测算法,例如 YOLO 这类可检测人员、车辆、动物等不同对象类别的 AI 模型。
用户自定义对象检测¶
ecu.test 中的对象检测使用对象检测器识别图像中的各个对象,并将其输出为 MatchList。为此,必须提供一个 Python 模块。要使其被识别为有效的 UserObjectDetector,必须满足以下要求:
该模块必须位于当前 Workspace(工作区)的
UserPyModules文件夹中。必须在脚本顶层定义变量 MODULE_TYPE = ‘ObjectDetector’。
必须实现一个名称任意、但满足以下接口的 ObjectDetector 类:
__init__
GetId
GetDocumentation
DetectObjects
GetAvailableParameters
GetParameter
SetParamter
不得存在两个具有相同 ID 的对象检测器。
有关该接口的更多细节,请参见下一节中的示例。
新增或修改后的对象检测器会在更新用户库后被 ecu.test 识别,此操作可在主菜单“Extras”中执行。
要通过内部 Multimedia API 使用用户自定义对象检测器,可以通过以下方式创建实例:
api.Multimedia.GetObjectDetector
返回对象上可以直接调用该接口中已实现的方法,例如
api.Multimedia.GetObjectDetector('ExampleObjectDetector').GetParamter('objects').
对象检测最小示例¶
自定义对象检测器的示例实现可在 ecu.test 安装目录中找到:
Templates/DefaultData/UserPyModules/UserObjectDetector/ExampleObjectDetector.py
ExampleObjectDetector.py 中的最小示例包含以下代码。有关各个方法的更多信息可参见文档字符串。
from __future__ import annotations
from typing import TYPE_CHECKING, Any
import numpy as np
if TYPE_CHECKING:
import numpy.typing as npt
MODULE_TYPE = 'ObjectDetector'
class ExampleObjectDetector:
def __init__(self):
"""
The object detector can be created when *GetObjectDetector(detectorId)* is called on
the internal multimedia API, where *detectorId* is the identifier returned by *GetId()*.
"""
self._params: dict[str, Any] = {
param['name']: param['default'] for param in self.GetAvailableParameters()
}
def DetectObjects(self, img: npt.NDArray) -> list[dict[str, int | float]]:
"""
Calls the detection algorithm of you object detector. This example detector creates
(pseudo-random) object matches in the image dimensions and returns it.
:param img: Numpy image (width x height x channels)
:return: A list of dictionaries that describe the (pseudo-random) objects in the image with
position, dimensions and confidence.
"""
objects = []
# retrieve image dimensions
height, width = img.shape[:2]
# create pseudo random number generator with fixed seed
gen = np.random.default_rng(seed=42)
# create objects at random positions. The number of objects is defined in the 'objects'
# parameter
for _ in range(self._params.get('objects', 10)):
objWidth = gen.random() * 100
objHeight = gen.random() * 100
x = gen.random() * (width - objWidth)
y = gen.random() * (height - objHeight)
objects.append(
{
'x': x,
'y': y,
'width': objWidth,
'height': objHeight,
'confidence': gen.random(),
}
)
return objects
@staticmethod
def GetDocumentation() -> str:
"""
Returns a short documentation for your ObjectDetector class.
:return: A documentation string.
"""
return (
'This is an example object detector that detects a parameterized number of objects at '
'random positions.'
)
@staticmethod
def GetId() -> str:
"""
The identifier of the object detector which is used to create the detector via the
multimedia api.
:return: Identifier of the object detector.
"""
return 'ExampleObjectDetector'
def GetAvailableParameters(self) -> list[dict[str, Any]]:
"""
Returns a list of all configureable parameters. Each parameter must be described by a unique
name and a default value. A short description is optional.
:return: List of all available paramters
"""
return [
{
'name': 'objects',
'default': 10,
'description': 'Number of objects to detect',
}
]
def GetParameter(self, name: str) -> Any:
"""
Get the current value of a given parameter.
:param name: Name of the parameter. Available parameters can be queried via
GetAvailableParameters() method.
:return: Value of the given parameter.
"""
return self._params.get(name, None)
def SetParameter(self, name: str, value: Any) -> None:
"""
Set a new value to the given parameter.
:param name: Name of the parameter. Available parameters can be queried via
GetAvailableParameters() method.
:param value: The new value of the given parameter.
"""
self._params[name] = value