Aiming at the problem of detection of floating targets by water surface cleaning robots, a method of fusion detection of 3D lidar point cloud data and visual information is proposed. First, the visual recognition part adopts CornerNet-Lite target detection network, through the training of a large number of samples to achieve the detection of floating objects on the water surface, and obtain the type and confidence of candidate targets. Then, through the calibration of the camera and lidar, the lidar three-dimensional point cloud data is projected onto the two-dimensional pixel plane, and the confidence of the lidar detection target is defined according to the concept of relative area size. Finally, adjust the confidence weight ratio of lidar and camera detection targets to form a new decision function, and determine whether the target is detected by comparing the size of the decision function with the set threshold. Experimental results show that this method has higher accuracy than CornerNet-Lite algorithm alone, eliminates the effects of water reflection and ripples, and reduces the false alarm rate.