LGMD (Lobula Giant Movement Detector) is part of a visual system of a locust, used to detect and evade approaching predators. Similar algorithm can be used in man-made systems like autonomous robots or safety systems in vehicles for detecting objects approaching on collision course. In this article usage of LGMD-based algorithms in road vehicles is investigated and a new solution is proposed. Video stream recorded from a moving vehicle inherently contains a lot of movement of the background due to vibrations and due to turning of the vehicle. This cause high output of LGMD and could trigger false alarms. The new approach, proposed in this article, enhances LGMD information and adds estimated expansion of objects in X and Y direction. Combination of all three sources of information gives a very good estimate of risk of collision. The new algorithm was developed using a reference set of test videos, where only one parameter was changed. Algorithm was tested with videos recorded from a moving vehicle in normal traffic and on the test ground, including real collisions in carton target at different speeds. Time between triggering of the alert and actual collision was measured.