Road vehicle detection based on multi-sensor heterogeneous fusion and adaptive Kalman filter in rainy and foggy scenarios

Traditional sensors are prone to sensitivity reduction in harsh weather conditions such as rain and fog. Therefore, this study designs a road vehicle detection model that combines multi-sensor heterogeneity and adaptive Kalman filter. This model integrates and analyzes data captured by LiDAR, millimeter wave radar, and optical cameras to enhance the sensor's perception ability in harsh environments. Among them, the adaptive Kalman filter can weight dynamic data streams, minimize the impact of noisy data, and maintain real-time response of real data. Experimental verification shows that under nominal conditions, the model achieved a vehicle detection accuracy of 95.5% with 2GB of memory consumption, while maintaining an accuracy of over 88.3% in severe weather conditions. The simulation results show that: (1) under thick fog conditions (visibility300m), the accuracy of model detection is 99.1%, and the RMSE is 0.6m. Under thick fog conditions, the response time of the model to moving objects is 0.34 seconds, and the detection accuracy can reach 93.5% at a speed of 90 kilometers per hour. The above data indicates that the model has good robustness and accuracy under adverse weather conditions. This research provides a perception solution for auto drive system in dynamic environment.