SafeVision:Smart Offline Animal and Fog Detection System

This paper presented a real-time road safety system designed to detect multiple hazards such as animals, pedestrians, and fog conditions using computer vision and deep learning techniques. The proposed system integrates object detection with environmental condition analysis to enhance driver awareness and reduce the likelihood of accidents, particularly in rural and highway environments.   The system was implemented using TensorFlow.js, enabling real-time processing directly within a web browser. This eliminates the need for high-end computational hardware and allows the system to function efficiently on standard devices. One of the key advantages of the proposed approach is its ability to operate in both offline and online modes. The offline mode ensures uninterrupted functionality in areas with limited or no internet connectivity, while the online mode provides flexibility in connected environments.   Experimental results demonstrated that the system is capable of accurately detecting animals and pedestrians using the COCO-SSD model, while also identifying foggy conditions through image processing techniques. The integration of these modules enables simultaneous detection of multiple hazards, which significantly improves overall system performance compared to existing solutions that focus on a single aspect of road safety.   Furthermore, the system was evaluated under different environmental conditions, including low-light and nighttime scenarios. The incorporation of night vision detection enhances the system’s reliability by enabling accurate detection even under reduced visibility conditions. The alert mechanism, which includes both visual and audio notifications, ensures that the driver receives timely warnings, thereby improving reaction time and situational awareness.   Despite its advantages, the system has certain limitations. The detection accuracy may be affected under extreme weather conditions such as heavy rain or dense fog. Additionally, variations in camera quality and resolution can influence system performance. The current implementation is also limited to predefined object classes and may require further training for broader real-world applications.   Future work can focus on enhancing the system by incorporating more advanced deep learning models such as YOLOv8 for improved accuracy and speed. The integration of additional sensors, including thermal imaging and LiDAR, can further improve detection in challenging conditions. Moreover, incorporating GPS and IoT-based communication systems can enable real-time alerts to nearby vehicles and emergency services, thereby extending the system’s functionality.   In conclusion, the proposed system provides a lightweight, cost-effective, and efficient solution for real-time road hazard detection. Its ability to operate under various environmental conditions, combined with offline capability and multi-hazard detection, makes it a practical and scalable approach for improving road safety in real-world scenarios.

[1] S. Sengan et al., “Real-Time Automatic Detection of Indian Roadway Animals Using Modified YOLOv3,” Journal of Intelligent Transportation Systems, 2021.   [2] N. Mamat et al., “Animal Intrusion Detection Using YOLOv5 Algorithm,” International Journal of Advanced Computer Science, 2022.   [3] K. Parkavi et al., “Enhancing Road Safety by Detecting Animals on Highways Using YOLOv5,” International Journal of Engineering Research, 2023.   [4] Brožovský et al., “Wild Animal Detection Using Thermal Imaging Cameras,” Sensors Journal, 2020.   [5] P. Balakrishnan et al., “Deep Learning-Based Animal Monitoring Using VGG16 and Deep SORT,” IEEE Access, 2021.   [6] Z. Liu et al., “Algorithm for Foggy Weather Target Detection Based on YOLOv5 in Complex Scenes,” 2022.   [7] Q. Wang et al., “Fog-YOLO: A Real-Time Object Detector for Foggy Autonomous Driving Scenarios,” 2023.   [8] Y. Qiu et al., “IDOD-YOLOV7: Image-Dehazing YOLOV7 for Object Detection in Low-Light Foggy Traffic Environments,” 2024.