Recognizing Suspicious Activities in Examination Scenario Using Machine Learning Techniques.

Monitoring the activities of examinees during examination is very challenging. The paper recognizes and classifies activities of examinees as suspicious or normal during examination using machine learning techniques. The processing and analysis of image data follows a typical sequence of distinct steps referred to as the vision pipeline. Data was acquired with a surveillance camera and frames extracted from the videos. Preprocessing activities include selecting the required frames from frame sequences, and cropping and segmenting foreground/background object. Video conversion to frame was accomplished with MATLAB scripts, while segmentation of image frames was achieved with GrabCut algorithm. Shape/pose features were extracted from objects using Histogram of Oriented Gradient (HOG) and Regionprop algorithms, and represented in feature vectors that were fed into Support Vector Machine (SVM) classifier. Holdout validation technique was used for the classifier training and tested from the given datasets. 70% of the dataset was used for training, while 30% was used for testing. The model gave an accuracy of 98.1% and 100%, respectively, for each examination scenario. The model accuracy was visualized in confusion matrix and the Receiver Operating Characteristic (ROC). MATLAB software was used as the simulation environment. The model demonstrated excellent performance, indicating that the system can adequately complement the efforts of invigilators in examination invigilation. [ABSTRACT FROM AUTHOR]