<![CDATA[The 2024 Indonesian SAR Robot Contest demands the ability of robots to differentiate between dummy dolls and victim dolls in emergency situations. This SAR robot has the main goal of rescuing victims and bringing them to a safe zone, so the author explores the implementation of object detection on SAR robots using ESP32-cam to detect victim dolls. The authors used the Edge Impulse platform, a TinyML platform, to train an object detection model using the Faster Objects, More Objects (FOMO) architecture. This model is optimized to run efficiently on resource-limited devices such as the ESP32-cam microcontroller. Training data was obtained by taking pictures of dummy dolls and victim dolls in various angles, lighting conditions and backgrounds using a camera from the ESP32-cam. The confusion matrix results from the model training process showed that the F1 score reached 100% and when testing the model, the object detection model was able to detect the victim doll with adequate accuracy, even though there were challenges such as variations in position and environmental conditions so the researchers used additional algorithms to increase detection accuracy. . The use of FOMO allows faster object detection and is able to detect more objects in one frame. This implementation shows great potential in the development of more efficient and autonomous SAR robots for rescue missions. These findings contribute to improving robotic technology, one of which is in SAR operations and provide a basis for further research in the application of object detection.]]>
Copyrights © 2024
