遮罩（Masks）

Securex NVR 中最常用的遮罩有两类：Motion Mask 与 Object Filter Mask。它们都是“微调工具”，不是把某块画面彻底从系统中删除。很多场景其实更适合使用 Zone。

Motion Mask

用于忽略某块区域的像素变化，例如时间戳、风吹树叶、天空或邻居电视。它只影响运动触发，不会把该区域从对象识别画面中抹掉。

Object Filter Mask

用于某一个对象类别的过滤。例如屋顶上经常误报 person，可以只对 person 设置过滤遮罩，而不影响 car。对象遮罩应尽量小，否则可能导致真实目标在经过边缘时被过滤。

示例

cameras:
  front:
    motion:
      mask:
        timestamp:
          friendly_name: Timestamp
          coordinates: "0.00,0.00,0.30,0.00,0.30,0.08,0.00,0.08"
    objects:
      filters:
        person:
          mask:
            roof:
              friendly_name: Roof false positives
              coordinates: "0.00,0.00,1.00,0.00,1.00,0.22,0.00,0.22"

什么时候不要使用 Mask

“只有进入车道才通知” → 使用 Zone + required_zones。
“不想记录后院” → 调整摄像头视角/隐私遮罩或录像策略，而不是依赖 Motion Mask。
“远处误报很多” → 先调对象尺寸/置信度过滤，再考虑小范围 Object Mask。

过度遮罩的风险

运动跟踪会利用运动区域来预测目标下一帧位置。大面积 motion mask 可能让目标跟踪不稳定；Object Mask 太大会直接丢掉合法目标。调整时一次只改一个多边形，并通过 Debug 验证。

Masks

Securex NVR commonly uses two mask types: motion masks and object filter masks. They are narrow tuning tools, not a way to completely remove a region from the system. Many use cases are better handled with zones.

Motion masks

Suppress pixel changes from nuisance regions such as timestamps, moving foliage, sky, or a TV. They affect motion triggering, not whether image pixels are available to object detection.

Object filter masks

Apply to a specific object class. For example, mask a roof only for person false positives without affecting cars. Keep these masks small.

Example

cameras:
  front:
    motion:
      mask:
        timestamp:
          friendly_name: Timestamp
          coordinates: "0.00,0.00,0.30,0.00,0.30,0.08,0.00,0.08"
    objects:
      filters:
        person:
          mask:
            roof:
              friendly_name: Roof false positives
              coordinates: "0.00,0.00,1.00,0.00,1.00,0.22,0.00,0.22"

When not to use a mask

“Notify only when entering the driveway” → use a zone plus required_zones.
“Never record the backyard” → use camera privacy/angle or recording policy.
“Many distant false positives” → tune size/confidence filters first.

Over-masking

Motion tracking uses motion regions to refine where an object will appear next. Large masks can make tracking unstable, while large object masks can discard real objects. Change one polygon at a time and verify in Debug.

Máscaras

Securex NVR usa principalmente motion masks y object filter masks. Son herramientas de ajuste fino, no una forma de eliminar por completo una parte de la imagen.

Motion Mask

Ignora cambios de píxeles de timestamp, árboles, cielo u otras zonas molestas. Solo afecta al disparo de movimiento.

Object Filter Mask

Se aplica a una clase concreta. Por ejemplo, puede filtrar falsos person en un tejado sin afectar car. Mantenga estas máscaras pequeñas.

Ejemplo

cameras:
  front:
    motion:
      mask:
        timestamp:
          friendly_name: Timestamp
          coordinates: "0.00,0.00,0.30,0.00,0.30,0.08,0.00,0.08"
    objects:
      filters:
        person:
          mask:
            roof:
              friendly_name: Roof false positives
              coordinates: "0.00,0.00,1.00,0.00,1.00,0.22,0.00,0.22"

Cuándo no usar máscara

“Avisar solo al entrar al driveway” → use Zone + required_zones.
“No grabar una zona” → use privacidad/encuadre o política de grabación.
“Falsos positivos lejanos” → ajuste tamaño y confianza primero.

Riesgo de exceso

Máscaras grandes pueden empeorar el seguimiento o descartar objetos reales. Cambie un polígono cada vez y verifique en Debug.