Desarrollo de un prototipo basado en Visión por Computadora para detectar la invasión del carril de buses en la ciudad de Loja.

Computer vision combines techniques of artificial intelligence, image processing, and deep learning to develop systems that can automatically analyze visual content and thus address real societal problems. In this context, in the city of Loja, it is evident that there is a significant number of vehicles invading lanes designated for public transportation, an infraction classified as fourth and sixth-class contravention according to the COIP. Therefore, the present research aims to develop a computer vision-based prototype to detect bus lane invasions. A development methodology was employed, encompassing the following stages: system implementation, image collection, data preparation, model training, and evaluation. The object detection model yielded promising results with an mAP50 of 92% across all classes and an mAP50-95 of 84%, indicating its capability to detect a wide range of objects with high precision under diverse conditions. Regarding invasion detection, a 94% precision was achieved with the system operating in real-time, concluding its high effectiveness for implementation. Moreover, it is noteworthy that this project provides invaluable support to transit authorities in identifying and penalizing drivers who invade the bus lane. However, it cannot operate autonomously; it requires the collaboration of an individual to verify and validate the data. Keywords: Object detector, COIP infractions, lane invasions, image processing, computer vision.

• Autor Principal: Román Carrión, Ibeth Katherine (author)
• Formato: bachelorThesis
• Idioma: spa
• Publicado: 2024
• Subjects: DETECTOR DE OBJETOS; INFRACCIONES COIP; INVASIONES DE CARRIL; PROCESAMIENTO DE IMÁGENES; VISIÓN POR COMPUTADORA
• Acceso en liña: https://dspace.unl.edu.ec/jspui/handle/123456789/31614