<![CDATA[Search And Rescue (SAR) robots are designed to locate and rescue victims from disaster sites to safe zones. In the 2024 Indonesian Search And Rescue Robot Contest (KRSRI), to earn points during rescue missions, the robot must identify victims among dummies with the victim oriented at a 45° angle and accurately place them in the safe zone. This research utilizes input from an Arducam camera and leverages the YOLOv4-Tiny Computer Vision algorithm, which offers reliable and adaptive detection and recognition capabilities under varying victim rescue conditions. The system outputs control commands for the robot's movement and manipulator arm. The final project successfully implemented the YOLOv4-Tiny model on a SAR robot, achieve real-time object detection at a minimum light intensity of 6 lux and a maximum distance of 60 cm. The system demonstrated a mAP of 99% and an IoU rate of 91.58%, with an average processing speed of 14.52 FPS. The success rate was 87.50% with an average time of 18.99 seconds for rescuing victims without dummies, and 70.83% with an average time of 46.78 seconds for rescuing victims among dummies. For victim placement, success rates and average times were as follows: 86.67% and 15.29 seconds for the gray safe zone, 93.33% and 15.23 seconds for the yellow safe zone, and 100% with 14.29 seconds for the marker safe zone. Given the high accuracy and speed, this algorithm is effective for scoring points in the Indonesian Search And Rescue Robot competition.]]>
