像素坐标在实际坐标的映射

侧视图

H：摄像头高度

Dmin: 图像底边距摄像头实际距离

Dmax: 图像顶边距摄像头实际距离

: 摄像头视角下边缘与竖直夹角

: 摄像头视角上边缘与下边缘夹角

: 目标点与摄像头视角下边缘夹角

俯视图

: 摄像头视角水平夹角

h: 图像像素高度

w: 图像像素宽度

x0: 目标点的像素x坐标

y0: 目标点的像素y坐标

x1: 目标点实际映射x坐标

y1: 目标点实际映射y坐标

B0: 摄像头视角最远出宽度/2

B1: 目标点所在位置x轴方向如图所示距离

 推理过程

使用

需要提前测得的量:

H, Dmin, Dmax, B0, h, w

输入：

x0, y0

输出：

x1, y1

python用

# 像素映射坐标
def get_location(x0, y0):
    H = 20
    Dmin = 20
    Dmax = 50
    B0 = 100
    beta = np.arctan(B0 / Dmax) * 2
    # beta = pi/3
    h = 640
    w = 640
    alpha = np.arctan(Dmin / H)
    theta = np.arctan(Dmax / H) - alpha
    delta = (h - y0) * theta / h
    y1 = H * math.tan(alpha + delta)
    B1 = (y1 + Dmin) * np.tan(beta / 2)
    x1 = 2 * B1 * (x0 - w / 2) / w
    return [x1, y1]


# frame = 相隔几帧计算速度
# fps = 一秒被分成几帧
# bias = 检测框抖动像素距离
# peak = 最大限速
def get_speed(id_dic, id, fps, bias=1, peak=120):
    frame = round(fps/4)
    if len(id_dic[id]) <= frame:
        speed = 0
    else:
        x1 = id_dic[id][-1][0]
        x2 = id_dic[id][-1 - frame][0]
        y1 = id_dic[id][-1][1]
        y2 = id_dic[id][-1 - frame][1]
        if abs(x1 - x2) <= bias and abs(y1 - y2) <= bias:
            speed = 0
        else:
            speed = math.sqrt(math.pow(x1 - x2, 2) + math.pow(y1 - y2, 2)) / (frame / fps)
            if speed > peak:
                speed = 0
    return round(speed)

# prediction
if len(outputs) > 0:
    for j, (output, conf) in enumerate(zip(outputs, confs)):

        bboxes = output[0:4]
        id = output[4]
        cls = output[5]
        x = (bboxes[0] + bboxes[2]) / 2
        y = bboxes[3]
        fps = vid_cap.get(cv2.CAP_PROP_FPS)
        if id_dic.__contains__(id):
            id_dic[id].append(get_location(x, y))
        else:
            id_dic[id] = [get_location(x, y)]
        speed = get_speed(id_dic, id, fps)

#待验证