Airtest图像识别原理

AirtestIDE 是一个跨平台的UI自动化测试编辑器,适用于游戏和App。

  • 自动化脚本录制、一键回放、报告查看,轻而易举实现自动化测试流程

  • 支持基于图像识别的Airtest框架,适用于所有Android和Windows游戏

  • 支持基于UI控件搜索的Poco框架,适用于Unity3d,Cocos2d与Android App

一句话总结:我们推出了两款基于Python的UI自动化测试框架Airtest(用截图写脚本)和Poco(用界面UI元素来写脚本),可以用我们提供的AirtestIDE来快速编写你的自动化测试脚本~

本文重点是针对Airtest中的图像识别进行代码走读,加深对图像识别原理的理解

1、准备工作,下载好源码 https://github.com/AirtestProject/Airtest

2、我们从最简单的touch方法入手,即为点击某个传入的图片,源码在api.py里面

@logwrap
def touch(v, times=1, **kwargs):
    """
    Perform the touch action on the device screen

    :param v: target to touch, either a Template instance or absolute coordinates (x, y)
    :param times: how many touches to be performed
    :param kwargs: platform specific `kwargs`, please refer to corresponding docs
    :return: finial position to be clicked
    :platforms: Android, Windows, iOS
    """
    if isinstance(v, Template):
        pos = loop_find(v, timeout=ST.FIND_TIMEOUT)
    else:
        try_log_screen()
        pos = v
    for _ in range(times):
        G.DEVICE.touch(pos, **kwargs)
        time.sleep(0.05)
    delay_after_operation()
    return pos

click = touch  # click is alias of touch

 这个函数执行点击操作的是 G.DEVICE.touch(pos, **kwargs)

而pos就是图片匹配返回的坐标位置,我们重点看loop_find这个函数

@logwrap
def loop_find(query, timeout=ST.FIND_TIMEOUT, threshold=None, interval=0.5, intervalfunc=None):
    """
    Search for image template in the screen until timeout

    Args:
        query: image template to be found in screenshot
        timeout: time interval how long to look for the image template
        threshold: default is None
        interval: sleep interval before next attempt to find the image template
        intervalfunc: function that is executed after unsuccessful attempt to find the image template

    Raises:
        TargetNotFoundError: when image template is not found in screenshot

    Returns:
        TargetNotFoundError if image template not found, otherwise returns the position where the image template has
        been found in screenshot

    """
    G.LOGGING.info("Try finding:\n%s", query)
    start_time = time.time()
    while True:
        screen = G.DEVICE.snapshot(filename=None)

        if screen is None:
            G.LOGGING.warning("Screen is None, may be locked")
        else:
            if threshold:
                query.threshold = threshold
            match_pos = query.match_in(screen)
            if match_pos:
                try_log_screen(screen)
                return match_pos

        if intervalfunc is not None:
            intervalfunc()

        # 超时则raise,未超时则进行下次循环:
        if (time.time() - start_time) > timeout:
            try_log_screen(screen)
            raise TargetNotFoundError('Picture %s not found in screen' % query)
        else:
            time.sleep(interval)

 首先会获取手机屏幕截图,然后对比脚本传入图片获取匹配上的位置  

match_pos = query.match_in(screen)

在cv.py 里面找到 Template类的 match_in方法

def match_in(self, screen):
    match_result = self._cv_match(screen)
    G.LOGGING.debug("match result: %s", match_result)
    if not match_result:
        return None
    focus_pos = TargetPos().getXY(match_result, self.target_pos)
    return focus_pos

重要的是 self._cv_match(screen)

    @logwrap
    def _cv_match(self, screen):
        # in case image file not exist in current directory:
        image = self._imread()
        image = self._resize_image(image, screen, ST.RESIZE_METHOD)
        ret = None
        for method in ST.CVSTRATEGY:
            if method == "tpl":
                ret = self._try_match(self._find_template, image, screen)
            elif method == "sift":
                ret = self._try_match(self._find_sift_in_predict_area, image, screen)
                if not ret:
                    ret = self._try_match(self._find_sift, image, screen)
            else:
                G.LOGGING.warning("Undefined method in CV_STRATEGY: %s", method)
            if ret:
                break
        return ret

 这里传入的图像需要进行缩放变化,写用例时候的截图进行变换后转换成跑用例时候的截图,以提高匹配成功率

image = self._resize_image(image, screen, ST.RESIZE_METHOD)

这里的匹配方法会遍历ST.CVSTRATEGY里面的方法,这个定义在Setting.py文件里面,默认是包含两种方法的

    CVSTRATEGY = ["tpl", "sift"]

如果某个方法匹配上了,就返回匹配结果,那么接下来就是重点搞清楚这几个方法是怎样实现的了。

_find_sift_in_predict_area也会调用到 _find_sift,那么接下重点就是分析这两个方法了

cv.py 中的  _find_template  _find_sift

    def _find_template(self, image, screen):
        return aircv.find_template(screen, image, threshold=self.threshold, rgb=self.rgb)

    def _find_sift(self, image, screen):
        return aircv.find_sift(screen, image, threshold=self.threshold, rgb=self.rgb)

 

3、先看

aircv.find_template 具体实现在 template.py 

def find_template(im_source, im_search, threshold=0.8, rgb=False):
    """函数功能:找到最优结果."""
    # 第一步:校验图像输入
    check_source_larger_than_search(im_source, im_search)
    # 第二步:计算模板匹配的结果矩阵res
    res = _get_template_result_matrix(im_source, im_search)
    # 第三步:依次获取匹配结果
    min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(res)
    h, w = im_search.shape[:2]
    # 求取可信度:
    confidence = _get_confidence_from_matrix(im_source, im_search, max_loc, max_val, w, h, rgb)
    # 求取识别位置: 目标中心 + 目标区域:
    middle_point, rectangle = _get_target_rectangle(max_loc, w, h)
    best_match = generate_result(middle_point, rectangle, confidence)
    LOGGING.debug("threshold=%s, result=%s" % (threshold, best_match))
    return best_match if confidence >= threshold else None

 重点在 _get_template_result_matrix

def _get_template_result_matrix(im_source, im_search):
    """求取模板匹配的结果矩阵."""
    # 灰度识别: cv2.matchTemplate( )只能处理灰度图片参数
    s_gray, i_gray = img_mat_rgb_2_gray(im_search), img_mat_rgb_2_gray(im_source)
    return cv2.matchTemplate(i_gray, s_gray, cv2.TM_CCOEFF_NORMED)

 这里可以看到,Airtest也没有自己研究一套很牛逼的算法,直接用的OpenCV的模板匹配方法

 

4、接着看另外一个方法

sift.py 

def find_sift(im_source, im_search, threshold=0.8, rgb=True, good_ratio=FILTER_RATIO):
    """基于sift进行图像识别,只筛选出最优区域."""
    # 第一步:检验图像是否正常:
    if not check_image_valid(im_source, im_search):
        return None

    # 第二步:获取特征点集并匹配出特征点对: 返回值 good, pypts, kp_sch, kp_src
    kp_sch, kp_src, good = _get_key_points(im_source, im_search, good_ratio)

    # 第三步:根据匹配点对(good),提取出来识别区域:
    if len(good) == 0:
        # 匹配点对为0,无法提取识别区域:
        return None
    elif len(good) == 1:
        # 匹配点对为1,可信度赋予设定值,并直接返回:
        return _handle_one_good_points(kp_src, good, threshold) if ONE_POINT_CONFI >= threshold else None
    elif len(good) == 2:
        # 匹配点对为2,根据点对求出目标区域,据此算出可信度:
        origin_result = _handle_two_good_points(im_source, im_search, kp_src, kp_sch, good)
        if isinstance(origin_result, dict):
            return origin_result if ONE_POINT_CONFI >= threshold else None
        else:
            middle_point, pypts, w_h_range = _handle_two_good_points(im_source, im_search, kp_src, kp_sch, good)
    elif len(good) == 3:
        # 匹配点对为3,取出点对,求出目标区域,据此算出可信度:
        origin_result = _handle_three_good_points(im_source, im_search, kp_src, kp_sch, good)
        if isinstance(origin_result, dict):
            return origin_result if ONE_POINT_CONFI >= threshold else None
        else:
            middle_point, pypts, w_h_range = _handle_three_good_points(im_source, im_search, kp_src, kp_sch, good)
    else:
        # 匹配点对 >= 4个,使用单矩阵映射求出目标区域,据此算出可信度:
        middle_point, pypts, w_h_range = _many_good_pts(im_source, im_search, kp_sch, kp_src, good)

    # 第四步:根据识别区域,求出结果可信度,并将结果进行返回:
    # 对识别结果进行合理性校验: 小于5个像素的,或者缩放超过5倍的,一律视为不合法直接raise.
    _target_error_check(w_h_range)
    # 将截图和识别结果缩放到大小一致,准备计算可信度
    x_min, x_max, y_min, y_max, w, h = w_h_range
    target_img = im_source[y_min:y_max, x_min:x_max]
    resize_img = cv2.resize(target_img, (w, h))
    confidence = _cal_sift_confidence(im_search, resize_img, rgb=rgb)

    best_match = generate_result(middle_point, pypts, confidence)
    print("[aircv][sift] threshold=%s, result=%s" % (threshold, best_match))
    return best_match if confidence >= threshold else None

 重点看如何找到特征点集

def _get_key_points(im_source, im_search, good_ratio):
    """根据传入图像,计算图像所有的特征点,并得到匹配特征点对."""
    # 准备工作: 初始化sift算子
    sift = _init_sift()
    # 第一步:获取特征点集,并匹配出特征点对: 返回值 good, pypts, kp_sch, kp_src
    kp_sch, des_sch = sift.detectAndCompute(im_search, None)
    kp_src, des_src = sift.detectAndCompute(im_source, None)
    # When apply knnmatch , make sure that number of features in both test and
    #       query image is greater than or equal to number of nearest neighbors in knn match.
    if len(kp_sch) < 2 or len(kp_src) < 2:
        raise NoSiftMatchPointError("Not enough feature points in input images !")

    # 匹配两个图片中的特征点集,k=2表示每个特征点取出2个最匹配的对应点:
    matches = FLANN.knnMatch(des_sch, des_src, k=2)
    good = []
    # good为特征点初选结果,剔除掉前两名匹配太接近的特征点,不是独特优秀的特征点直接筛除(多目标识别情况直接不适用)
    for m, n in matches:
        if m.distance < good_ratio * n.distance:
            good.append(m)
    # good点需要去除重复的部分,(设定源图像不能有重复点)去重时将src图像中的重复点找出即可
    # 去重策略:允许搜索图像对源图像的特征点映射一对多,不允许多对一重复(即不能源图像上一个点对应搜索图像的多个点)
    good_diff, diff_good_point = [], [[]]
    for m in good:
        diff_point = [int(kp_src[m.trainIdx].pt[0]), int(kp_src[m.trainIdx].pt[1])]
        if diff_point not in diff_good_point:
            good_diff.append(m)
            diff_good_point.append(diff_point)
    good = good_diff

    return kp_sch, kp_src, good

 至于这个sift是什么对象

def _init_sift():
    """Make sure that there is SIFT module in OpenCV."""
    if cv2.__version__.startswith("3."):
        # OpenCV3.x, sift is in contrib module, you need to compile it seperately.
        try:
            sift = cv2.xfeatures2d.SIFT_create(edgeThreshold=10)
        except:
            print("to use SIFT, you should build contrib with opencv3.0")
            raise NoSIFTModuleError("There is no SIFT module in your OpenCV environment !")
    else:
        # OpenCV2.x, just use it.
        sift = cv2.SIFT(edgeThreshold=10)

    return sift

可以看到,用到的也是OpenCV的方法,如果是OpenCV3则查找图像特征点集的方法就是

cv2.xfeatures2d.SIFT_create(edgeThreshold=10).detectAndCompute() 

 

5.总结下来,最终用到的就是OpenCV的两个方法,模版匹配和特征匹配

1.模板匹配  cv2.mathTemplate

2.特征匹配 cv2.FlannBasedMatcher(index_params,search_params).knnMatch(des1,des2,k=2)

哪个优先匹配上了,就直接返回结果

 

6、总结

  1. 图像识别,对不能用ui控件定位的地方的,使用图像识别来定位,对一些自定义控件、H5、小程序、游戏,都可以支持;
  2. 支持多个终端,使用图像识别的话可以一套代码兼容android和ios哦,用ui控件定位的话需要兼容一下。
  3. 缺点:对于背景透明的按钮或者控件,识别难度加大

你可能感兴趣的:(Airtest图像识别原理)