Robust Hybrid CNN-BiGRU based Surveillance System for Wild Animal Detection and Tracking to Enhance Rural Safety and Forestry

Wild animal detection and tracking are crucial for rural safety and forest conservation, enabling proactive monitoring and conflict prevention, thus promoting sustainable ecosystem management. However, image-based detection faces challenges such as occlusion, illumination variations, diverse animal appearances, complex backgrounds, and limited annotated datasets, adversely affecting model accuracy. Existing literature indicates that standalone CNNs struggle with temporal dependencies, while VGG-19 with Bi-LSTM incurs high computational costs and slow inference. Additionally, both models are susceptible to environmental noise, reducing detection robustness. This research proposes a hybrid CNN-BiGRU framework for wildlife surveillance, where CNN extracts spatial features, and BiGRU processes temporal dependencies across video frames, enhancing detection continuity. The proposed Hybrid CNN-BiGRU outperforms than existing standalone CNN and GRU models, achieving a superior accuracy of 98.18% in dynamic, real-world conditions. Results demonstrate its efficacy in mitigating human-wildlife conflicts, improving conservation strategies, and offering a resilient, high-performance detection system adaptable to challenging environmental conditions.