Immunization injection is of ten require d to prevent and control the diseases in pig farming. However, current needle injection still relies on manual labor in pig farms, leading to serious challenges, such as high risk of needle breakage, severe cross-infection, low efficiency, time-consuming, and labor-intensive operation, as well as missed or incorrect injections. In this study, a control system was proposed for vision-based positioning and motion in a needle-free injection robot under confined stall breeding scenarios. According to the workflow of swine immunization, the key components were selected to integrate the navigation system of the robot. Its parameters were then determined to improve the needle-free injection module. The motion range of the robotic arm was simulated using MATLAB. The rotation angles of its six axes were constrained to prevent collision between the robotic arm and confined stall. A precise localization was proposed for the needle-free injection site on pigs. The YOLOv8n framework was improved to solve the missed and false detections caused by blurred features of the tail root and susceptibility to dirt interference. The improved network was used to extract the tail root region from pig images. A point set of the hip muscle injection area was constructed for vertical injection. The least squares method (LSM) was applied to fit the surface point set, enabling accurate calculation of the injection point and posture. A control system was developed to accurately track the injection point in the needle-free injection robot, due to the random movement of pigs during operation. According to feeding system, the injection was divided into two phases: a pre-injection and an injection phase. In the pre-injection phase, the tail root position was continuously detected to perform visual servoing control of robotic arm, thereby tracking the injection point in real time. In the injection phase, the robot moved into the injection point, and then performed the injection, according to the last detected pose. The critical distance of phase transition was determined to be 15 cm using depth camera and triangulation. The robot operating system (ROS) was used to integrate navigation, visual detection, and robotic arm control algorithms, enabling the robot to follow a predetermined trajectory and then perform needle-free injection on pigs. Experiments were conducted at a pig farm in Qinhuangdao City, Hebei Province, China. The experimental results showed that the tail root detection algorithm achieved an accuracy of 95.8%, a recall of 93.5%, and a mean average precision (mAP) of 97.1%. The overall injection success rate of the robot was 93.3%, of which 95.2% were vertical injections. The maximum injection deviations in the X, Y, and Z axes were 2.9, 3.7, and 1.9 cm, respectively, corresponding to average deviations of 1.30, 1.91, and 0.59 cm. The injection accuracy was fully met the requirements of swine immunization. A closed-loop control system was realized from chassis navigation, visual recognition, dynamic tracking to precise triggering for the needle-free injection in real scenarios. The findings can also offer the technical and engineering reference for the large-scale intelligent farming.