A vehicle detection apparatus includes a lamp candidate extraction unit that extracts a pixel region that may correspond to a tail lamp of a vehicle from pixel regions that an integration processing unit creates by extracting and integrating pixels of an image as a lamp candidate and a grouping unit that regroups groups containing the lamp candidate of the groups generated by grouping position data detected by a position detection unit and then regroups all groups. In the regrouping processing, a threshold used for regrouping groups containing the lamp candidate is set easier for regrouping than a threshold used for subsequently regrouping all groups.