Search for: All records

The increasing number of online courses and programs available worldwide has elevated the importance of reliable online exam proctoring. The typical proctoring process relies on webcam surveillance. However, this traditional method of proctoring is vulnerable to numerous types of face occlusions used for religious reasons or otherwise. We present a robust biometric authentication framework that combines advanced eye and face recognition, resulting in a much better live proctoring system that is further augmented by including fingerprinting. We have used cutting-edge deep learning techniques, specifically the Siamese network for fingerprint analysis and a ResNet-based eye recognition model tested with and without Gabor filters. Furthermore, our system offers much better performance compared to previously existing models. Notably, our system maintains high accuracy, approximately 98.04% for custom eye recognition, 99.01% for publicly available labeled faces dataset, 82% for niqab dataset and 87.04% for publicly available fingerprint dataset. Moreover, our model demonstrates a 10–20% improvement in face recognition under occlusion. Our solution is highly effective for not only online proctoring but also allows use in other similar situations, such as employee authentication for remote presence verification. Our system supports scenarios involving partial occlusions, such as masks and sunglasses, to full occlusions with veils, without requiring any additional hardware. 

Pedestrian regulation can prevent crowd accidents and improve crowd safety in densely populated areas. Recent studies use mobile robots to regulate pedestrian flows for desired collective motion through the effect of passive human-robot interaction (HRI). This paper formulates a robot motion planning problem for the optimization of two merging pedestrian flows moving through a bottleneck exit. To address the challenge of feature representation of complex human motion dynamics under the effect of HRI, we propose using a deep neural network to model the mapping from the image input of pedestrian environments to the output of robot motion decisions. The robot motion planner is trained end-to-end using a deep reinforcement learning algorithm, which avoids hand-crafted feature detection and extraction, thus improving the learning capability for complex dynamic problems. Our proposed approach is validated in simulated experiments, and its performance is evaluated. The results demonstrate that the robot is able to find optimal motion decisions that maximize the pedestrian outflow in different flow conditions, and the pedestrian-accumulated outflow increases significantly compared to cases without robot regulation and with random robot motion.