Real-time wildlife monitoring on edge devices faces challenges such as power constraints, limited bandwidth, and intermittent connectivity, especially in remote European habitats. Object detection-only systems transmit every detected frame and use frame-based counting, resulting in high data overhead, even when the same animal remains within the camera’s field of view for extended periods. Counting is further complicated by occlusions, low resolution, and varying lighting. We propose a lightweight tracking system based on Local Binary Pattern (LBP) similarity, integrated with a custom YOLOv8m model. The tracker leverages Euclidean and cosine similarity for robust re-identification under partial occlusion and illumination changes. Optimized for low-power edge deployment, our system significantly reduces redundant frame transmission and improves counting accuracy. It achieves up to 99.67% bandwidth savings and 99.9% power reduction, enabling scalable, efficient wildlife monitoring. We validate our system on embedded platforms, demonstrating high performance in constrained environments.