Enhancing Pedestrian Pose Detection through YOLO Deep Learning Techniques

The rapidly growing field of computer vision holds the promise of transforming numerous aspects of existence, enhancing both quality of life and safety measures. Computer vision also has many applications, one of which is pedestrian detection, which is used in many areas of life such as autonomous driving, monitoring systems, and traffic control. Pedestrian detection algorithms often struggle with many problems such as size variation, varying lighting conditions, or excessive computational execution time, which makes them unsuitable for real-time systems. To address these limitations, this paper proposes an approach that integrates filters with a deep learning model, YOLOv8x. A dataset containing 478 images of walking and sitting poses was utilized to detect pedestrian pose detection. Our model achieved detection fitness and precision of ${9 0 \%}$ for images without pre-processing techniques, and $95 \%$ precision for images after applying a Gaussian filter and with $83 \%$ fitness, and after applying a Laplacian Filter, a precision of ${8 3 \%}$ and fitness of $52 \%$ was obtained. We found that if accuracy is important, applying filters is the best option, and if speed and efficiency are important then pose detection without pre-processing will be efficient.