In this paper, we present a multi-object detection approach leveraging the enhanced YOLOP algorithm, which is designed to address the challenge of detection of road vehicles, lane lines, and drivable areas in autonomous driving systems. Traditional single-task learning methods face challenges such as high computational resource consumption and insufficient model generalization ability when dealing with complex scenes. To this end, we design a multitask learning network architecture that simultaneously handles vehicle detection, lane line and drivable area segmentation tasks by sharing a feature extraction layer. The network architecture consists of three parts: an encoder, a task decoupled decoder and a multi-objective loss function. The encoder adopts a hybrid CNN-Transformer structure, the decoder designs specialized detection and segmentation heads for different tasks, and end-to-end training is performed by a multitask joint loss function. The results show that the algorithm performs well on the BDD100K dataset and can significantly improve the detection accuracy and the segmentation effect while ensuring real-time performance.