AniLarm: AI for Small Animal Detection Under Cars Using Thermal Camera

Urban safety concerns, particularly regarding small animals under cars, present a significant challenge. To address this, a research prototype named AniLarm was developed, applying offline AI (no internet dependency) to detect small animals using thermal imaging and machine learning. The system integrates a Seek Thermal Compact Camera with a Raspberry Pi for real-time classification and detection. AniLarm uses the EfficientNet model, outperforming YOLOv5 with a test accuracy of 97.83 % on a balanced dataset. The system evolved through multiple iterations, from LED-based alerts to a web application, ultimately adopting an architecture with a Raspberry Pi in a protective case, monitor, and Bluetooth speaker for auditory alerts. Thermal images are processed end-to-end in approximately 8 seconds, with results delivered via audio notifications. The prototype was tested in both controlled environments and real-world scenarios, effectively distinguishing animals from nonanimal objects after addressing misclassifications and latency. This research demonstrates the feasibility of integrating an offline trained machine learning model with an embedded system and a vision sensor to improve safety in urban applications.