A Novel Approach of Object Detection in Video Surveillance Systems Using DeepSORT and CapsNet-Based Access Control Procedure

In video surveillance, video sequences are analyzed with computer vision technology requiring more data storage. Computer vision technology has the challenging task of improving the detection performance of object movement through the analysis of video sequences, where the binary mask on each frame is considered for detection analysis. More applications like video surveillance, human behavior identification, industrial inspection, traffic control, and action recognition support this object detection task. Moving cameras are mostly used in various applications and the challenging factors on detection performance include sensor noise, scene occlusion, diverse illustrations, and pose variation. Henceforth, a new deep learning-based model is implemented for object detection to resolve all such issues. Firstly, access control is used for accessing the video surveillance system, in which the Capsule Network (CapsNet) is used for providing the results as either unauthorized or not. After authorized access, the required videos are fetched. Since it holds with low light, occlusion, and motion blur, the video enhancement is achieved by Retinex. Finally, the tracking of objects is carried out by employing the DeepSORT model. Lastly, the proposed model is executed and validated with certain measures. These findings demonstrate that the proposed work attains higher performance in detecting the objects from the video source effectively than other traditional approaches.