Abstract: Vehicle-mounted millimeter wave radar possesses the advantage of all-weather operation; however, due to the scarcity of semantic information, it is arduous to identify the Target category from the radar echo signal. To address the aforementioned issues, a feature-enhanced target detection network(F-TDNet) based on camera assistance is devised. Firstly, a feature enhancement module, namely Conv3d BatchNorm ReLU(CBR), was proposed to enhance the feature extraction capability of the model. Secondly, convolutional autoencoder is employed to guarantee the comprehensiveness of radar semantic information acquisition. Subsequently, an squeeze-and-excitation networks(SE) attention mechanism featuring fewer parameters and less computation is added to enhance the refinement ability of feature extraction. The experimental results indicate that the average accuracy attains 92.50% and the recall rate amounts to 95.86% under the circumstances of unstable ambient light and variable target speed. In comparison with other approaches, the proposed method and the improved network exhibit an outstanding detection effect. The method proposed in this paper reduces the reliance of the automatic driving system on the camera and possesses the characteristics of all-weather detection and stable performance, which is conducive to enhancing the safety of the auxiliary driving system.