Optimal deep learning based object detection for pedestrian and anomaly recognition model

Object detection, specially designed for anomaly and pedestrian recognition, epitomizes the forefront of computer vision (CV) advancement. By harnessing the abilities of deep learning (DL), this complex system adeptly detects pedestrians within varied environments, using neural networks (NNs) and convolutional NNs (CNN) for accurate detection. This technology excels in accurately detecting pedestrians and anomalies in various environments by using NNs and CNNs. This study introduces an Optimal DL based Object Detection for Pedestrian and Anomaly Recognition (ODLOD-PAR) technique. The purpose of the ODLOD-PAR technique is to detect and classify the presence of pedestrians and anomalies using DL models. To accomplish this, the ODLOD-PAR technique applies image pre-processing using bilateral filtering (BF) based noise removal and dynamic histogram equalization (DHE) based contrast enhancement approach. Besides, the ODLOD-PAR technique uses YOLOv5m model for object detection process. Moreover, the YOLOv5m involves CSPBep backbone, CSPRepBiFPAN neck, and EffiDeHead head modules. To enhance the detection outcomes of the YOLOv5m technique, the hyperparameter tuning process takes place using RMSProp optimizer. The performance evaluation of the ODLOD-PAR approach undergoes utilizing benchmark UCSD anomaly detection dataset. The experimental outcomes highlighted that the ODLOD-PAR method obtains optimum solutions under other approaches with maximum AUC score of 99.05%.