Abstract: For the problems of complex detection process, many influencing factors and low detection accuracy of traditional in-car DUI detection methods, an improved tabu search (TS) - BP neural network fusion in-car DUI detection method is proposed considering the driver and passenger positions and open window areas. Firstly, the TS algorithm is improved by setting different domain ranges in each search phase; secondly, the initial weights and thresholds of the BP neural network are optimized using the improved algorithm to improve the efficiency and accuracy of the detection model and avoid falling into local optimal solutions; finally, the data from multiple sensors are fused to achieve high-precision in-car DUI detection. The experiments show that the proposed detection method improves 65.78%, 88.20% and 58.38% of the model average absolute error, mean square error, and average absolute percentage error indexes, respectively, and the convergence speed and performance and robustness of the model are significantly improved compared with the traditional method. This study can provide a reference for in-vehicle portable DUI high-precision detection.