The social welfare institution known as LPKS plays a crucial role in providing protection and guidance to children, though its implementation remains suboptimal. This discussion focuses on the LPKS strategy in shaping the social behavior of children with legal issues and the factors hindering its effectiveness. This research employs a qualitative descriptive method with a phenomenological approach, involving direct field observations, interviews, and documentation. The findings regarding the LPKS strategy in shaping the social behavior of children with legal problems include several aspects. Firstly, there is a normative role. Secondly, LPKS performs a practical role by addressing various needs such as mentoring, psychological therapy, spiritual or religious therapy, arts training activities, and legal protection throughout the legal process. Additionally, it provides educational guidance focusing on the developmental and mental aspects of children. Finally, the ideal role is to enhance LPKS performance. The primary inhibiting factor affecting the LPKS strategy in shaping the social behavior of children with legal problems is budget limitations. These limitations arise due to the substantial budget required to cover a large area, while existing resources remain insufficient.

Utilization of computer vision can be applied in various aspects of daily life, reducing dependence on human labor. One of its implementations is in industry, such as in the production process of motorized vehicles, to sort or classify parts or goods. The computer vision process involves many stages, such as image capture, image processing, image analysis, image recognition, and decision-making. In the automotive industry, computer vision has been used in autonomous or driverless electric vehicles, as well as in creating intelligent transportation systems. To detect objects in real-time, one of the options that can be used is to use the YOLO algorithm, which can detect objects in one stage with predictions of bounding boxes and class probabilities simultaneously. However, although YOLO has good performance, the architecture has some drawbacks, such as complexity and complicated hyperparameter congurations. To remedy this, the Adam optimization algorithm was introduced, which combines the momentum and RMSprop algorithms to adjust the learning rate adaptively and provide faster convergence in model training. This is evidenced by an increase in the value of mAP on Yolov5. These results prove that the Yolov5 method with Adam`s optimization is better than the Yolov5 method without optimization.