A set of Computer Vision algorithms was validated through a facial recognition-based access control system designed for indoor environments, motivated by the necessity for contactless authentication and the imperative to surmount the limitations inherent to traditional cards and passwords. The research, characterized as applied and quantitative and utilizing the hypothetical-deductive method, performed performance evaluations based on objective metrics within controlled scenarios. The methodology included a Systematic Literature Review (SLR) in which nine databases were consulted, resulting in 381 initial records, which were subsequently reduced to 45 eligible primary studies following the application of exclusion criteria. Within the literature, precision ranged from 78.4% to 100% (mean: 96%) and accuracy from 80% to 100% (mean: 94%), signaling maturity in controlled environments, despite persistent gaps regarding standardized reporting. Regarding the prototype, implemented in Python with facial detectors and extractors, trials conducted with 6 cameras of distinct configurations demonstrated a strong dependence on capture conditions. specifically, average accuracy increased from 65.7% at 110 lux to 82.1% at 515 lux. Conversely, long-distance capture (350 cm) caused a drop in the average from 92.7% to 32.2%, except when utilizing dedicated optical zoom, thereby reinforcing the importance of facial scale on the sensor. Furthermore, the utilization of a pre-processing technique enhanced accuracy, ensuring greater reliability in facial recognition tasks. Multiple additional scenarios were evaluated, such as spoofing attempts, user motion, the use of accessories, partial facial occlusion, and night vision capture. This work presents a set of techniques operationalized through a web-based system and attests to the applicability of this type of solution in both controlled and real-world scenarios. By considering a range of acquisition devices under distinct conditions, this study provides the necessary tools to facilitate replicability.