Spatial and Temporal Characteristic Analysis based Long Short-Term Memory for Detection of Sensor Fault in Autonomous Vehicles

Abstract

The artificial intelligence required to create self-directed automobiles relies heavily on the capability of precisely perceiving the environment around oneself. Most self-driving automobiles include several detectors, which work together to form a multi-source perception of the surroundings. Extended use of a system that drives  autonomously will introduce a variety of worldwide and local failure indications due to the extreme sensitivity of the instruments involved to ambient or environmental situations. These failure indications pose significant risks to the technique’s security. The paper presents a real-time information synthesis system incorporating techniques for identifying flaws and accepting faults. The compact connection can be recognized if the qualities mentioned above are provided, and the input information properties may be retrieved in real-time. One way to use the newly introduced method for assessing device reliability is to compute the detectors’ worldwide and local degrees of trustworthiness. In order to ensure the precision and dependability of information combination, problem data is filtered out, and monitor duplication is used to assess both the worldwide and local assurance levels of data from sensors at the moment. The chronological and geographic association of data from sensors allows for this. Experimental findings show that the network’s algorithms can outperform current techniques in terms of both rapidity and precision and can pinpoint the object’s location even when specific sensors are blurry or broken. This research established that the proposed hybrid structure benefits autonomous vehicles’ real-time reliability and speed.