引導式測驗 · 中等 等級

AI分析身體攝像機錄像測驗

測試人工智慧隨身攝影機分析:搜尋與識別、上下文、編輯、本地驗證、來源視訊審查、存取和保留。

問題 1 的 8

A clothing search returns a clip showing a person in a red jacket. What does that result establish?

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. A clothing search returns a clip showing a person in a red jacket. What does that result establish?
    • The person committed the event under investigation
    • The clip contains visual features matching the query, not a verified identity
    • The person is uniquely identified by clothing
    • The search returned every clip containing that jacket
  2. Why should reviewers inspect footage before and after an AI-selected segment?
    • The model cannot store a timestamp
    • A longer clip always has higher image resolution
    • The system requires the reviewer to accept every result
    • The surrounding events can change how the isolated clip should be interpreted
  3. A body-camera search returns no results for a brief gesture. Which explanation remains possible?
    • Frame sampling or occlusion may have caused the system to miss it
    • The gesture definitely did not occur
    • The video is inadmissible
    • A no-result search verifies every frame manually
  4. Which evidence should accompany a consequential AI search result?
    • Only a screenshot of the search result
    • A confidence score without source video
    • The original clip, timestamps, query details, and system information needed to reproduce review
    • A narrative written by the search model
  5. What should an agency measure when testing an AI redaction tool?
    • Only the number of minutes processed
    • Missed faces or plates as well as false detections across representative footage
    • Only user satisfaction
    • The length of the vendor manual
  6. A generated summary says an officer issued a command, but the audio is unclear. What is the right response?
    • Verify against the original recording and label uncertainty rather than adopting the summary as fact
    • Treat the generated sentence as an independent witness
    • Remove the video once the summary is saved
    • Use the summary because it is concise
  7. Why may a vendor benchmark fail to predict performance in a city’s deployment?
    • All video models perform identically across environments
    • Benchmarks always use every local camera
    • Local testing changes the model’s legal status
    • Local cameras, lighting, scenes, and recording conditions may differ from the test data
  8. Which policy most directly supports accountable use of footage search?
    • Allow any employee to search any footage without logging
    • Keep every derivative permanently by default
    • Limit searches by purpose and access, keep logs, and review uses against retention and disclosure rules
    • Let the vendor decide which investigations qualify