Real-time Bangla License Plate Recognition System for Low Resource Video-based Applications

Intelligent Video Surveillance Systems for Vehicle Identification Based on Multinet Architecture

“…On the other hand, there is another very interesting approach that consists of deploying these systems for low-resource devices in real time, emulating real environments. For example, in [41], the authors run their tool in a CPU-based system with 8 GB of RAM, and obtain a precision of 66.1% with MobileNet SSDv2 [42] with a 27.2 FPS rate; or in [43], with metrics of detection of 90% and a recognition rate of 98.73% with just Raspberry Pi3B+ as hardware support. Going further, we can find Android-based systems such as [44] or [45], which are designed to be deployed in mobile phones and operate in real time.…”

“…Under these circumstances, it may be better to opt for less precise but less resourceconsuming systems. For example, some ALPR systems are ready to operate through CPU instead of GPU, although they sacrifice precision; such as [41] or [43]. It is important to mention that in critical scenarios, there will always be a human controller behind the system, so these tools may deliver more false positives rather than false negatives (and the human controller will discriminate if it is right or not).…”

Deep Learning for Robust Vehicle Identification

“…Por otro lado, existe otro enfoque muy interesante que consiste en desplegar estos sistemas para dispositivos de bajos recursos en tiempo real, emulando entornos operativos reales. Por ejemplo, en [72], los autores ejecutan su herramienta en un sistema basado en CPU con 8 GB de RAM, y obtienen una precisión del 66,1% con MobileNet SSDv2 [73] con una tasa de 27,2 FPS; o en [74], con métricas de detección del 90% y una tasa de reconocimiento del 98,73% con sólo Raspberry Pi3B+ como soporte hardware.…”