With the introduction of deep learning technology in recent years, target detection algorithms for autonomous vehicle have made significant progress. This paper analyzes and organizes the traditional object detection algorithms and deep learning object detection algorithms currently applied in autonomous driving from the perspective of the development of object detection technology, analyzes milestone detectors, network structures and the latest detection methods, and explores the development direction of target detection technology.

To address the scarcity of multi-source heterogeneous data and insufficient scenario adaptability in current perception algorithm training and testing of autonomous driving, a typical scenario-based multimodal perception dataset is constructed. It contains 10 specific typical scenario segments, covering multimodal sensor data from LiDAR, cameras, and 4D millimeter-wave radar. The dateset provides annotation information for six categories of targets and offers detailed descriptions of data acquisition device configurations, including sensor parameters, calibration data, and a time synchronization processing scheme. By delivering scenario-specific driving context, the constructed dataset enhances perception accuracy in complex environments, thereby improving the safety and reliability of autonomous driving systems.

To achieve more efficient detection of small traffic sign targets under complex urban street background conditions, this paper proposes an improved YOLOv5s algorithm. This enhancement is achieved by incorporating a Convolution Block Attention Module (CBAM) Spatial Channel Attention Mechanism, an Adaptive Spatial Feature Fusion (ASFF) module, and an improved loss function for detection boxes. The validation results on the TT100K traffic sign dataset demonstrate that the proposed algorithm achieves a mean Average Precision (mAP) of 84.5% in traffic sign recognition.

In the process of driving a vehicle, the complex and changing environment inside the vehicle, the change of lighting conditions and the diversity of drivers’ behavioral postures affect the detection and recognition of abnormal driver behavior. To address this issue, this paper proposes a driver abnormal driving behavior detection algorithm based on contrast learning. The paper firstly considers driver’s driving behavior detection as a binary classification task, and utilizes a contrast learning approach to compare driver’s normal driving with abnormal driving samples and to improve the performance of the model by contrasting loss functions. Secondly, the depth images right ahead and above the driver serves as inputs to solve the problems of complex in-vehicle environment to change the light intensity and blind spots in viewpoint by providing the depth information of the driver. Finally, 3D convolution is introduced in the lightweight network MobileNetV2, and the operation of channel blending is added to the convolution layer of each bottleneck structure to improve the accuracy of recognition. Test results show that accuracy of the proposed algorithm reaches 94.18% in the Driver’s Abnormality Detection (DAD) dataset and ROC AUC reaches 0.962, which shows the effectiveness of the algorithm in driver’s abnormal behavior detection.

In order to analyze the effects of In-Vehicle Traffic Lights (IVTL) on driving behavioral characteristics, a driver data collection system is designed in an environment with obstructed line-of-sight. Fifty human subjects between the ages of 20 and 40 are recruited for the driving simulator test, and vehicle and driver status data are collected. After preprocessing, statistical methods are used to analyze the correlation indicators. The results show that compared with driving without IVTL, the average speed of vehicles equipped with IVTL increase significantly. In the condition of traffic lights obstructed by large vehicle, the distance between following vehicle and leading vehicle is reduced, which potentially improves road traffic efficiency, and IVTL can alleviate the stress of drivers brought by the abnormal road environments.

To address the challenge where the traditional Adaptive Cruise Control (ACC) are limited to maintain low-speed following when encountering a low-speed vehicle in front, an ACC control system with lane change function is developed. Firstly the system subdivides the driving modes into three types: cruise control, cruise following and lane change cruise, and a multi-mode switching strategy based on speed dissatisfaction is formulated to flexibly respond to different driving scenarios and driver’s needs. On the basis of the existing adaptive cruise system, the active lane change function is added, and the quintic polynomial is used to accurately plan the lane change trajectory, and then the lane change cruise trajectory tracking controller is constructed based on the Model Predictive Control (MPC) algorithm. Finally, the controller is verified based on MATLAB/Simulink/CarSim. The simulation results show that the proposed strategy meets the requirements of active lane change in line with the driver’s intention.