Aiming at the problem of map offset caused by deceleration zone or pothole zone encountered by vehicles in automatic parking scenarios, and the problem of poor real-time performance and low accuracy of the system caused by changing dynamic environment, a variable sliding window look-around SLAM algorithm for automatic parking is proposed. Firstly, IMU is used to calibrate the real-time attitude of the collected panoramic image to improve the accuracy of mapping. Secondly, combined with the advantages of multi-sensor fusion, the IMU and odometer data are fused to estimate the vehicle pose. Finally, the variable sliding window algorithm is used to accelerate the back-end optimization and improve the real-time performance and accuracy of the system. The simulation test results show that the method solves the problem of map offset in the deceleration zone or the pit zone, and the efficiency and real-time performance are improved by 31.35% and 25.06% respectively in the sparse feature environment. The real vehicle test results show that the method can achieve the positioning accuracy with an average error of 0.039 m, which provides safety guarantee for parking.