YOLOV4 Based Blind Assistant System for Real-Time Object Detection
FRANCIS SHAMILI S*1, PRATHYUSH N P*2, KENES K ROBY*3 PRADEEP S*4
*1Assistant professor, Department of CSE, Dhanalakshmi Srinivasan Engineering College (Autonomous), Perambalur, Tamil Nadu, India.
*2,3,4UG Students, Department of CSE, Dhanalakshmi Srinivasan Engineering College (Autonomous), Perambalur, Tamil Nadu, India.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract – This project focuses on object detection using the YOLOv4-tiny model, a lightweight version of the YOLO (You Only Look Once) algorithm designed for real-time object detection. The model is loaded with pre-trained weights and a configuration file, making it capable of detecting various objects from a webcam feed. Once an object is detected, the system identifies the class, evaluates the confidence of the detection, and calculates the bounding box coordinates to highlight the object in the image. The system applies Non-Maximum Suppression (NMS) to remove overlapping bounding boxes and retain only the most relevant ones. In this implementation, the program captures video frames from a webcam, pre-processes them to a format suitable for the YOLO model, and passes them through the neural network to generate predictions. These predictions are then analyzed to identify the objects in the frame, and relevant details such as the object’s label and confidence score are extracted. If an object belongs to a specified class (e.g., "elephant," "bird," "horse," or "zebra"), the system triggers an HTTP request to send this information to the Blynk IoT platform for remote monitoring. The integration with Blynk IoT allows for real-time monitoring and remote alerts. By sending the detected object’s label to Blynk, the system facilitates quick action based on the objects being tracked. This setup could be used for various applications, including surveillance, automated tracking systems, and environments where real-time detection and remote reporting are essential. Additionally, the use of the YOLOv4-tiny model ensures that the object detection process is both fast and efficient, making it suitable for applications requiring low-latency responses.
