Intelligent Wild Animal Detection And Adaptive Risk Alert System Using Yolov8 And Deepsort

Abstract

Human–wildlife Conflict Is A Growing Global Concern As Expanding Human Settlements Encroach Upon Natural Habitats. Traditional Wildlife Monitoring Methods—manual Patrols And Passive Camera Traps—suffer From Delayed Response, Limited Scalability, And An Inability To Assess Threat Severity In Real Time. This Paper Presents An Intelligent Wild Animal Detection And Adaptive Risk Alert System That Integrates YOLOv8-based Real-time Object Detection, DeepSORT Multi-object Tracking With Persistent Unique Identification, Trajectory-based Behavior Classification (roaming, Approaching, Aggressive), A Novel Five-factor Composite Risk Scoring Engine With Sigmoid Normalization, Species-aware Dynamic Geofencing, Automated Sound Deterrents, And Multichannel Alert Dispatch Via SMS And WhatsApp. The System Is Orchestrated Through A Modular Python Pipeline And Visualized On A Live FastAPI Web Dashboard With WebSocketdriven Video Streaming. Experimental Evaluation On Wildlife Surveillance Footage Demonstrates A Mean Average Precision (mAP@0.5) Of 87.3% For Detection, 94.6% Tracking Consistency (MOTA), And Sub-200ms End-to-end Latency Per Frame On Consumer-grade Hardware. The Adaptive Risk Formula— Rfinal= σ α•Sd•ψ(b)+β•(1−dnorm)•v+γ•log2(n+1)+ —enables Context-sensitive Threat Es- Calation, Reducing False-positive Alerts By 41% Compared To Static Threshold Baselines. The Platform’s Modular Architecture And YAML-driven Configuration Support Rapid Deployment At Wildlife Corridors, Forest Perimeters, And Agricultural Buffer Zones.

Keywords