In the production process of wide and thick plates, the rough rolling process requires the control of intersecting conical rollers to achieve the adjustment of slab length and width direction. This process heavily relies on manual operation, and the rhythm cannot be effectively controlled, which hinders the intelligent transformation process of the production line. The automatic steel transfer system for rough rolling slab, designed by combining visual inspection and automatic control technology, can effectively solve this problem. Video surveillance cameras are installed at the entrance and exit of the rough rolling mill to capture the status of the slab in the steel transfer roller area. The PIDNet model with an improved defogging module is used to extract the foreground contour of the slab, and the slab rotation angle is tracked in realtime through a combination angle measurement method. During the process, strategies such as safety limit, position optimization, speed regulation, and over rotation correction are integrated to jointly optimize the control of the steel conversion process, ensuring the stability and safety of the steel conversion. Autonomous learning of manual experience improves the efficiency of steel conversion. The application results show that the system can accurately measure the angle of the slab and achieve automatic steel conversion function, which can replace manual operation, save energy consumption, and achieve the goal of intelligent production.