Evaluting the Performance of the Yolov7 Algorithm :A Comparative Study of iPhone and Samsung Smartphones Under Varying Lighting Conditions

This study evaluates the performance of the YOLOv7 algorithm for real-time object detection, emphasizing the impact of smartphone hardware capabilities (iPhone vs. Samsung) and environmental lighting conditions (day vs. night). Through extensive testing on diverse datasets, including urban scenes from Ajdabiya city, YOLOv7 demonstrated robust accuracy for high-contrast, well-represented objects such as cars (up to 0.96 accuracy) and appliances (e.g., microwave: 0.91). However, significant variability was observed in detecting occluded or small-scale objects (e.g., people: 0.33–0.88; plant pot: 0.28) and underrepresented classes (e.g., fire extinguishers: undetected). Hardware-specific disparities emerged: iPhones outperformed Samsung devices in low-light scenarios (person detection: 0.88 vs. 0.85), while Samsung exhibited superior dynamic range for trucks (0.90 vs. 0.89). Environmental factors, such as glare and overexposure, further exacerbated detection inconsistencies, particularly for traffic lights (nighttime range: 0.34–0.52). The study identifies critical gaps in YOLOv7’s generalizability, including sensitivity to dataset bias and environmental conditions, and underscores the need for hardware-aware preprocessing and dataset diversification. Future research should prioritize adaptive thresholding techniques and context-specific calibration to enhance reliability in real-world applications such as urban surveillance and autonomous systems.