Data collection and data processing determines the success of an incident detection system. The former provides the traffic data types and affects the configuration and performance of the resulting data processing algorithms. There have been a countless number of methods and algorithms developed for incident detection. They mainly rely on inductive loop detectors and loop emulators. Point data sources are notoriously unreliable and do not provide the best traffic data for determining real-time traffic conditions. The results of previous research indicate that no significant improvement on incident detection performance will be obtained until new types of traffic data collection technologies with high accuracy, reliability and coverage are used. This paper identifies and defines a new classification system based on three distinct incident detection types: roadway-based, probe-based and driver-based, in terms of their different sensor data (collection) types. The results of our nationwide survey on traffic management centers and traffic operations centers about their current and soon to be implemented traffic sensors and incident detection strategies suggest a movement from the traditionally used traffic sensing techniques to driver-based witness report. No current traffic sensing technology can provide the richness of incident detection and description and the coverage on roadway systems like driver-based incident detection techniques. In this paper, driver-based techniques, their effectiveness and performances are reviewed, identified and assessed. The implementation and benefit-cost issues are also analyzed and briefly compared with other incident detection approaches.