Optimisation of highway vehicle occlusion recognition based on attention and multitasking approach

The complex vehicle occlusion scenes on highways pose great challenges for vehicle detection and recognition. To raise the precision and robustness of vehicle detection, this study extracts vehicle features through keypoint detection technology and unbiased coordinate system transformation, combines multi-scale attention mechanism to process multiple tasks, and accurately identifies occluded vehicles. An occlusion recognition model that integrates attention mechanisms and multi-task learning is proposed. The experiment findings indicate that the model achieved an F1 value of 92.82%, a mean square error of 0.01, and a mean absolute error of 0.02 on the COCO dataset. Contrary to other mainstream algorithm models, the new model has the highest vehicle color detection precision of 94.56%, the highest vehicle type detection accuracy of 89.06%, and the shortest detection time of 0.21 seconds. From this, the detection precision of the model has significantly improved in complex scenes, proving its superior performance in identifying occluded vehicles. It is suitable for intelligent transportation applications on highways and provides reliable support for future highway vehicle occlusion recognition.