Development of a camera-based illumination controller for face detection, tracking and recognition using computational intelligence
Date of Publication
Doctor of Philosophy in Electronics and Communications Engineering
Electrical and Computer Engineering
Gokongwei College of Engineering
Electronics And Communications Engg
Elmer P. Dadios
Defense Panel Chair
Defense Panel Member
Laurence Gan Lim
Illumination is crucial in human activities and in machine vision applications. For indoor surveillance applications, Infrared (IR) Light Emitting Diodes (LEDs) are the common means of providing illumination to the camera to cause no discomfort to human occupants. While IR provides non-obtrusive illumination for the camera, the same energy consumed does not provide the illumination to indoor spaces of the building. This is important if the premises where the camera is installed is not connected to the main power source or electric grid but derives energy from renewable sources. This work details the development of an illumination controller integrated to a vision system that performs human object detection and face recognition under an energy-constrained environment.
The Camera-Based Illumination using Computational Intelligence Controller (CBIC-CInt) provides automatic adjustment of illumination in order to detect human object, and to recognize human person in a video stream with the lowest power use. Using Computer Vision algorithms, CBIC-CInt can detect the presence of a human person in a video stream, perform tracking, and recognize the person using face detection and recognition. The system provides the optimum illumination level to perform the detection, tracking and face recognition operations. Using visible Light Emitting Diodes as source of illumination, CBIC – CInt provides illumination both for the proper operation of the camera and human personnel monitoring the premises where the system is installed. This feature is significant in energy-constrained surveillance applications or where there is no power source derived from the electric grid.
Results showed that the developed computer vision system inferred the distance between the human body and the image sensor within 20 feet at a measured illumination of 7.2 lux. This is equivalent to 80% duty cycle PWM activation of the LED lighting system. Face detection and recognition is successfully carried out up to a distance of 8 feet at 12.7 lux illumination equivalent to 30% duty cycle. Body detection and tracking is successfully carried out within 8 feet to 18 feet distance from the image sensor with LED lighting system activated between 60% to 80% duty cycle. Face detection, tracking and recognition accuracy is 60% while body detection and tracking accuracy is 64%. Power consumption of the system under operation is at 5.8 watts. Results also showed that the performance of the system in both algorithms is severely affected fluorescent lighting but performs very well under natural lighting condition. In this research, a dimmable LED lighting system was implemented by fabricating an LED luminaire and driven by an Arduino microcontroller through PWM. The increment of illumination change is limited by the hardware interface between the vision system and the PWM controller of the lighting system which is fixed to 5% duty cycle. Photometric analysis of the luminaire was carried out to characterize its illumination capabilities.
Different object detection and recognition methodologies were explored until suitable platform for low-power and portable applications was identified. Object recognition using an FPGA platform was considered as well as face recognition using Deep Learning paradigm were examined. However, these methods require powerful computing platforms not suitable for portable and off-grid applications. An embedded system based on the Raspberry Pi was configured to implement, test and validate the operation of the body recognition and face recognition algorithms using the OpenCV library and the Python programming language. The embedded hardware has modest computing capability from its quad core processors, with a Linux-based operating system that can be installed and configured on a 32 GB secondary, non-volatile storage. Human body detection using the Moments of Gradients algorithm was implemented on the embedded platform together with the face detection algorithm using the haar cascades for frontal face was used to detect human faces on real-time video, and the two Takagi-Sugeno fuzzy logic controllers for illumination control.
Integration of the computer vision and the fuzzy logic controllers in the RPi embedded hardware, and its interfacing to the Arduino-based LED lighting subsystem, allowed the computer vision algorithms to track and infer the distance between the object and the image sensor. Both the human body detection and tracking, and the face detection and recognition algorithms inferred distances of body and face from the camera and feed these data to the two Takagi-Sugeno type fuzzy logic controllers to adjust the level of illumination correspondingly. The computer vision system infers the distance of the tracked objects based on the coordinates of the bounding box enclosing the detected object. It is also noted that the width of the bounding box can also be used to infer the distance.
The novelty of this research is based on the utilization of the computer vision algorithms to:1. Infer the distance of detected occupants in a video image and convert it as fuzzy inputs to a fuzzy controller to adjust the illumination intensity; 2. Incrementally adjust the intensity of illumination to a level that will enable detection and recognition of object in a video stream in real-time; 3. Provides illumination both for the computer vision system and human occupant resulting to minimized energy consumption; 4. Detects occupancy of the premises and recognize identity of the occupant when occupancy is within 8 feet from the image sensor.
Lighting; Human face recognition (Computer science)
Llorente, C. A. (2018). Development of a camera-based illumination controller for face detection, tracking and recognition using computational intelligence. Retrieved from https://animorepository.dlsu.edu.ph/etd_doctoral/1443
Upload Full Text