step (1): acquiring a point cloud and an image data frame which are spatiotemporal aligned in real-time through a solid-state laser radar and a camera, segmenting all target instances in the image data frame, and defining an image region where a target instance whose advanced semantic category is a networked vehicle is located as an advanced semantic constraint region;

step (2): using an affine transformation matrix from the point cloud to the image data frame to map all point clouds in a point cloud frame to a corresponding image, retaining mapping points falling in the advanced semantic constraint region, and discarding mapping points falling in other regions;

step (3): for a networked vehicle target instance in the advanced semantic constraint region, collecting all mapping points falling within its instance region, dividing a plurality of depth intervals at equal intervals according to the depth values of the mapping points, and designing an independent non-uniform increment sampling method to perform independent increment sampling on each depth interval to generate virtual mapping points, wherein a sampling rate is set according to a point density of each depth interval, a higher sampling rate is set for a depth interval with a low point density, and a highest sampling number for each interval is limited, in such a manner that sparse and missing regions of the mapping points are partially filled, and invalid sampling is avoided; and

step (4): reversely mapping the virtual mapping points to an original point cloud space to merge the virtual mapping points with an original point cloud, and then detecting networked vehicle targets and its' center point coordinate by using the merged point cloud to generate position estimation information of the networked vehicle.