TBC-K-Means based Co-Located Object Recognition with Co-Located Object Status Identification Framework Using MAX-GRU

Abstract

In the application of detached object recognition in public places like railway terminals, the recognition of the co-located objects in the video is a more vital process. Nevertheless, owing to the occurrence of multiple co-located object instances, the analysis of the status of the co-located object in the video is a challenging process. Hence, for solving this issue, this paper proposes the Min-Max Distance based K-Means (MMD-K-Means)-centric co-located object recognition with object status identification. Primarily, the input video from the railway is converted to frames. Subsequently, it was improved using Contrast Limited Adaptive Histogram Equalization (CLAHE). Next, Tukey’s Bi-weight Correlation-based Byte Tacking (TBC-BT) and MMD-K-Means clustering are done for the detection and tracking of moving and non-moving objects. Subsequently, the Cyclic Neighbor-based Connected Component Analysis (CN-CCA) process was done from the static and moving object-detected frames for the main and co-located object labeling. Next, it executed the patch extraction for the separate analysis of each instance. At last, the Maxout-based Gated Recurrent Unit (Max-GRU) determined the object status in CN-CCA processed frame with the estimated distance between objects and extracted features from the static objects. The proposed system was then experimentally examined and validated in contrast to the standard methods. The proposed MMD-K-Means achieved a co-located object identification rate of 97.92% in 1184 milliseconds. Next, the Max-GRU achieved 98.13% identification accuracy, and it also achieved excellent results for other performance parameters. The proposed system’s performance is experimentally proved with several performance metrics.