A Hybrid Deep Learning and Metaheuristic Optimization for Multi-Object Detection and Spatio-Temporal Analysis in Autonomous Navigation for Indian Traffic Systems

Abstract

Autonomous navigation in Indian traffic system is one among the trending technology in Intelligent Transportation System (ITS) but still it consists of certain challenge such as high density, heterogeneous vehicles, unpredictable pedestrian behavior and complex road conditions. So that it becomes very essential to create a model with safe and efficient navigation with reliable multi-object detection particularly in a developing country like India which is the second largest populated country in the world. For that purpose, in this article o novel a IMDHDO (Improved Multi-Object Detection with Hybrid DL Methods and Optimization) model is developed which perform effective multi-object detection and spatio-temporal anomaly detection. The core modules of this proposed model are dataset creation, Object detection using hybrid HRU-Net and LuNet architecture, YOLOv12 Network, spatial-temporal anomaly detection and hybrid optimization. In the hybrid HRU-Net and LuNet model, convolution-based process is incorporated which leads to perform effective target tracking using a Kalman filter. Through the spatio-temporal analysis, the vehicle features are effectively extracted even in the irregular traffic patterns. Finally, a hybrid optimization process is developed with the combination of Bat Optimization Algorithm (BOA) and Pigeon Inspired Optimization (PIO) which is used to attain independence in autonomous vehicle navigation. This hybrid model is effective enough to achieve maximum accuracy in analysing the real-world traffic data under low light or densely populated area. The experimental demonstration of this model is carried out in Python and the calculation matrix are various loss analysis, flexible and targeted navigation, multi-object tracking accuracy and so on and the obtained results making it suitable for real-world environments.