A dual-camera monitor was proposed and a detection algorithm was designed for abnormal cheating behavior in exams， which detected four common exam abnormal behaviors of candidates， such as： carrying prohibited items， surrogate exam-taker， speaking during the exam， and drifting of candidates' attention. Firstly， the attention‑based YOLOv5s algorithm was used to detect prohibited items in exams. Then， Dlib‑based facial recognition was applied to confirm the candidates' identity information. Finally， lip movement detection and head pose estimation were utilized to detect abnormal behavior of candidates during the exam. The use of dual cameras could reduce blind spots in the field of view to prevent candidates from using cheating tools. The experiment results showed that the proposed algorithm had good detection accuracy and real-time performance.