引導式測驗 · 中等 等級

物件追蹤的卡爾曼濾波器測驗

測試卡爾曼預測和校正、狀態和噪音假設、多目標追蹤的限制以及實際的序列級評估。

問題 1 的 8

卡爾曼濾波器在估計移動物體時保持什麼?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. 卡爾曼濾波器在估計移動物體時保持什麼?
    • A complete image detector and appearance database
    • A verified assignment between every detection and track
    • A state estimate and an estimate of its uncertainty
    • Only the object’s class label
  2. 預測步驟期間發生哪些計算?
    • It computes a new object class from the image
    • It advances the current estimate through the motion model before the next measurement update
    • It replaces the observation by a detector result
    • It assigns a detection to one of several tracks
  3. 當新的噪音觀測值到達時,校正步驟會做什麼?
    • Advance the state without using any observation
    • Detect a new object directly from pixels
    • Delete every track that was not observed in one frame
    • Combine it with the predicted state using their uncertainty
  4. 檢測器遺失一幀。過濾器能起到什麼作用?
    • A model-based predicted state with increased uncertainty until another observation arrives
    • Proof that the object remained visible
    • A new image detection made from no pixels
    • A guaranteed correct identity match
  5. 哪一個元件決定哪一個偵測屬於多目標系統中的哪一個軌道?
    • The object detector’s class vocabulary
    • A separate data-association process
    • The Kalman state prediction alone
    • The process-noise covariance by itself
  6. What does process-noise covariance Q represent in the linear prediction?
    • Uncertainty in the detector’s class names only
    • The number of objects in the scene
    • The measured object position itself
    • Uncertainty in how well the motion model describes state changes
  7. Which evaluation reveals whether a Kalman-based tracker works on video sequences?
    • Measure localization error, continuity, identity switches, and latency on representative sequences
    • Report only detector accuracy on still images
    • Tune Q and R until one sample looks smooth
    • Count state variables without testing track outcomes
  8. When might a basic linear Kalman filter be a poor fit?
    • When the filter outputs a covariance matrix
    • When a detector supplies a bounding box
    • When the system dynamics are strongly nonlinear or the noise assumptions are unsuitable
    • When measurements contain any noise at all