A multimodal perception method for distance and speed of front vehicle integrating vehicle image features was proposed for front vehicle detection and operational state perception in driving scenarios. The position features of vehicles in images were detected by an improved SW-YOLOv8n model, and the relative lateral and longitudinal distances to the front vehicle were calculated using geometric algorithms. A feature extraction network was designed to extract vehicle features, where image feature vectors were fused through serial concatenation, and a neural network for vehicle distance measurement was established. The multi-feature fusion module was integrated with the distance measurement neural network to construct an end-to-end front vehicle distance perception model and a vehicle tracking-based speed estimation model, which synchronously output precise distance estimations and stable speed tracking results. Experimental results demonstrated that on the test dataset, the SW-YOLOv8n model achieved improvements of 1.6 percentage points in mAP50 and 2.3 percentage points in mAP50?95 compared to the baseline YOLOv8n, while maintaining a detection speed of 260.11 frames per second. Within a lateral range of 9.5 m and a longitudinal range of 50 m, under unobstructed conditions, the preceding vehicle distance perception model exhibited an average relative error of 1.87% between predicted and actual distances, while under occluded conditions, the average relative error was 2.02%. The speed measurement results of the tracking-based model exhibited significant stability, confirming the method’s effectiveness for front vehicle distance and speed perception tasks.