Abstract: With the development of autonomous driving, multi-sensor fusion is widely-used. CLOCs is a 3D target detection network based on post-fusion, but it has poor detection performance for obscured objects. To address this problem, this paper proposes a 3D target detection network fusing binocular ranging and GRU (Gated Recurrent Unit). It introduced binocular ranging in the 2D target detection network to correlate 2D and 3D depth information based on the intersection over union fused 3D and 2D of CLOCs network, and added GRU network after convolution to capture the temporal data dependencies. The kitti dataset was used for validation. The experimental results show that the detection accuracy is improved.