Aiming at the adjustment and optimization of enterprise productivity structure, this paper proposes a research on jBZ-23-07 rotor-type oil pump intelligent assembly system based on machine vision. According to the requirements, the overall scheme design is formulated and the intelligent assembly system is decomposed into visual system, upper computer software system and other parts according to the design method of "from the whole to the part". Halcon was used to complete camera calibration and distortion correction, process part images, extract relevant features Based on shapeless multi-template matching method, and obtain attitude adjustment parameters. The image processing time of a single part is 130 ms and the position Angle error is 0.01°. The accuracy, stability and sensitivity of the algorithm recognition are verified by multi-method and multi-angle experiments. Finally, the average success rate of algorithm identification was 97.60%, 100.00 % and 98.04 %, respectively, meeting the overall design requirements and actual production requirements.