In this paper, a collision avoidance system is presented to detect red light running and warn nearby vehicles and pedestrians in real time in order to prevent possible accidents. No complex infrastructure-based solution such as those based on radars or cameras is here required. Instead, a new solution based on smartphones carried by drivers and pedestrians is proposed so that it is the device inside the vehicle violating a traffic light, the one that self-reports the offence in order to generate alerts and warn nearby vehicles and pedestrians to prevent accidents. The proposal could also be used by road authorities to collect data on traffic lights that are most frequently violated in order to define an action plan to investigate causes and look for solutions. It includes a classifier for learning and estimating driver behaviour based on collected data, which is used to predict whether he/she is about to run a red light or detect whether that has already happened. In the first case, the system broadcasts warnings directly to close vehicles and pedestrians through Wi-Fi, while in the second case, the proposal warns vehicles and pedestrians in the neighbourhood through a server. The solution also includes a prioritization system based on changing traffic lights at intersections according to the needs and characteristics of the traffic at all times, giving the top priority to emergency vehicles. Furthermore, the proposal involves the use of cryptographic schemes to protect authenticity and integrity of messages sent from traffic lights, smartphones and servers, and privacy and anonymity to promote the use of the system. A beta version with some parts of the proposal has been implemented and the obtained results are promising.