This research investigates the development of emotion detection based on the YOLOv9 model, emphasizing hyperparameter tuning to maximize performance. Most classical emotion recognition techniques are based on facial keypoints, which might lead to a decrease in accuracy as a result of occlusions and changes in face orientation. We introduce an object detection model that detects emotional states without relying on keypoints. Using the FER2013 dataset, consisting of grayscale images labeled into seven different emotions, we used Multi-task Cascaded Convolutional Networks (MTCNNs) for efficient face detection and bounding box retrieval. We tested it in varied scenarios, such as consumer-oriented product testing (toys and food) and entertainment (video games and films) for the 12–16 age group. The hyperparameter optimization of YOLOv9 attained a training accuracy of 92%, a 10% increase from conventional methods. These results highlight the potential of object detection algorithms to improve emotion recognition technologies and significantly contribute to artificial intelligence applications that involve comprehending human emotions. This research opens the door to future work in mental state prediction and interactive product development.