1. Introduction

In recent life, object detection and tracking system had a wide variety of applications. Such applications may include video surveillance, vision-based control, human-computer interfaces, medical imaging, augmented reality, animation, security, and robot vision. The machine vision system uses images and videos to detect, classify, and track objects or events to understand the real-world scene. Image processing is an active field of interest for researchers. The field of improvement, innovation, development, and modification in image processing is constantly expanding ^[1,2]. The most popular image-processing task are face recognition and tracking systems. In such systems, the object can be detected, and tracked, and a laser pointer can be aimed toward an object, to be used later in many applications, such as security and robotics. Recently, several problems resulted in these systems, such as the accuracy of detection, response time, and the obstacles that reduce or obscure the target under investigation.

2. Related Work

Many researchers have focused on object detection and tracking. A. H. Miry ^[3] implemented a real-time facial recognition system using webcam hardware as an input device and an Arduino microcontroller as an output device. The wavelet transform extracted the features with principal component analysis that provided data abundance. The results showed that the wavelet maximizes the distance for people outside the database while minimizing the distance for people in the database and found that the format 160*120 with wavelet transform was the best format for recognition than other formats.

R. Kiran et al. ^[4] reported an algorithm for face recognition using image processing while the output was pin state processing of an Arduino board with an ATmega328P controller by tracking a human face. When a board receives a character from MATLAB, a digital output pin or a board sets HIGH or LOW accordingly. LEDs are connected to these digital output ports and by receiving the signals from MATLAB, they changed their states from on to off and off to on.

V. S. Pamulapati et al. ^[5] explained the working of the Viola-Jones algorithm for face tracking in real-time. The main objective of their project was to detect a human face in every frame of a video coming from a web camera. The captured image was processed using the viola-jones algorithm using MATLAB to detect the faces and send signals to the Arduino board to control the camera movement using two servo motors. One servo was used for horizontal rotation and another for vertical rotation. The face was tracked actively and maintained in the frame. The proposed system for tracking faces using the Viola-Jones algorithm in MATLAB was superior the current method. Using an adaptive boosting algorithm helped increase the accuracy and efficiency of the proposed method.

P. Vignan ^[6] introduced a tracking system that used an inexpensive imaging device with two servo motors to control the pan and tilt of a camera for face tracking by capturing vital facial features. Through the results, the system was tested and the success rate was accurate (95.7\%).

M. Mira et al. ^[7] suggested the application of security systems to vehicles equipped with alarm devices on a large scale to detect the presence of persons. In such a system the security level was very low because the system could not distinguish between the owner and a thief. The trick was to match the texture of the facial curves with the facial data stored in the database. The vehicle can only be operated when the system detects the owner's face entered into the system database. If the driver's face not in the system database, it is not recognized, and the tool will automatically sound an alarm and turn off the car ignition. Based on data from the experiment, it was concluded that the voltage generated for some samples entered into the database were only a few differences in voltage values. On the other hand, compared to several samples not included in the database, there was a big difference in the voltage value. The voltage stability on each part tended to be stable, differing only by 0.5 volts.

B. Nethravathi et al. ^[8] proposed an effective design by implementing a home security system. The proposed work consisted of home security with real-time access and control via a mobile phone and an Arduino as a controller. A security system with a webcam was deployed at the entrance to the house. This was connected to the Arduino. The GSM module was used to send alert messages. The faces were recognized by the EHD algorithm using MATLAB. TLS was a very secure protocol, and could be posted to ensure no security was pierced. The system accuracy decreased when the number of different faces increased.

E. Ramkumar et al. ^[9] explained using the Viola-Jones database for the biometric authentication of individuals such that images are preserved during processing. The saved information was worked to recognize faces, and an impression signal was given to the console if the information matched. MATLAB software was used to give away control signals to the actuator, which was used to open and close the door. The image entrance was fed by a digital camera and the image was processed inside MATLAB. The output proposed was sent to the external controller interfaced with MATLAB. After implementing the system, it was possible to use the Arduino controller instead of the Raspberry pi, because it is less expensive and can accommodate larger datasets, all of which will improve the system efficiency.

T. A. Salih et al. ^[10] proposed a wireless camera system to detect objects in the field of view of a robot. The principal component analysis (PCA) algorithm and filters were used to implement and demonstrate the image operation. In such a system, the applied camera identified objects with changing backlight conditions such as a fire inside a building. The system was tested using the MATLAB environment, and the experimental performance demonstrated the efficiency and robustness of the proposed system.

The originality of the presented work focused on the accuracy of the measurements for servo used motor in terms of distances and angle of view according to the input information sent from the Arduino board.

3. Proposed Scheme

The following methods were adopted to test the performance of the designed system.

3.1 Cascade Object Detector

The cascade object detector system comes with several pertained classifiers for detecting frontal faces, profile faces, noses, eyes, and the upper human body. On the other hand, these classifiers are not always sufficient for a particular application ^[11]. A cascade of classifiers is a degenerated decision tree comprised of stages of increasing complexity, with the first stage training a classifier to detect most objects of interest and then triggering the evaluation of the second stage classifier, which has also been adjusted to achieve a high detection rate. Several integrated (nested) layers, each containing a boosted classifier, comprise a cascade of boosted classifiers. The cascade function as a single classifier combines the results of the previous steps. The reason behind this method is that the majority of the sub-windows within a single visual frame are negative. While it is uncommon for positive-sub-windows to travel through the stages. The cascade can speed up processing the number of sub-windows considerably using this method. The initial weak classifiers attempt to reject many negative sub-windows as possible, with great computation resources spent on the positive sub-windows. Sub-windows were more difficult in the first cascade stages. The Cascade detector used only one of three feature types (Hog-features, Haars features, and Local Binary Patterns features (LBP)). The default feature for the cascade object detector was Hog-features. The detection process takes the shape of a degenerate decision tree as depicted in Fig. 1 ^[12,13].

Fig. 1. Schematic depiction of the cascade detection[14].

4. Tools and Methodology

Arduino is an electronic development board consisting of an open-source electronic circuit with a microcontroller ``ATmega328'' which can be connected to different programs on a PC and relies on its programming on the open-source programming language processing, and codes. Arduino software is similar to C language and is one of the easiest programming languages for writing microcontroller programs. Fig. 2 depicts the main components used in the current system. A servo motor (MG995) with a 180-degree rotation angle was used. In addition to a laser pointer and human face image to detect and track. The algorithm was implemented using a Lenovo laptop camera (core i7 and resolution: 1280 X 720) with a (0$^{\circ}$ - 77$^{\circ}$) angle for the camera field of view.

The working principle of this system was as follows, it sends a notification signal to the Arduino board in which the object position relative to the x and y-axes are included when the laptop camera detects a face or any moving object. Subsequently, the Arduino board converts the signal to the servomotor, where a laser pointer is installed. As a result, the signal becomes 'one' when the moving object is detected correctly to activate the servomotor and direct it towards the object. Otherwise, it is 'zero' when no object is detected. Fig. 3 presents the whole face detection and tracking process.

Fig. 4 presents the Arduino program of the Arduino board is connected with the Matlab program.

The following expressions can be used to measure the accuracy of the servo motor ^[16]:

where ${\Delta}$x is the difference in meter between the specified initial angle and the range (0$^{\circ}$ - 77$^{\circ}$) in the real world, $\overline{\mathrm{x}}$ is the average error, $\overline{\mathrm{xs}}$ is an average error for square distances, $\sigma$ err is the standard deviation, Sth is an image scale (distance in pixels over distance in meters), ${\Delta}$Ө is the difference in pixel between the specified initial angle and the range (0$^{\circ}$ - 77$^{\circ}$), and C is object’s distance in meters in the real world.

Fig. 2. (a) Used tools; (b) Experimental setup.

Fig. 3. Block Diagram for the target detection and tracking system.

Fig. 4. Algorithm for detecting and tracking objects.

5. Results and Discussion

The working system was tested for several object distances ranging from one meter to 5m with increasing 0.5m in each step. Fig. 5 depicts the application of the system with increasing distances for some selected cases. The first, second, and third columns in the figure represents the results when the horizontal orientation angle (Ө-initial) equals zero, 40$^{\circ}$, and 77$^{\circ}$ respectively.

For the three orientation angles (i.e., Ө = 0$^{\circ}$, 40$^{\circ}$, and 77$^{\circ}$), the average error, image scale, and standard deviation are calculated for each object distance. Table 1 lists the results.

The highest accuracy of servo motor calibration had an average error equal to zero for all distances (Table 1).

Such calibration is executed by examining the servo motor when returning to its initial horizontal orientation angle (i.e., Ө = 0$^{\circ}$, 40$^{\circ}$, and 77$^{\circ}$) through several training steps. A reverse relationship resulted in Sth variation with increasing object distances. Figs. 6 and 7 show the parameter variations with increasing distances and theta respectively. Figs. 6(a), 6(b), 7(a), 7(b) show that the average error and standard deviation are very small and tend to zero with increasing distances and theta respectively which indicate the highest level of system efficiency. As mentioned earlier, a reverse relationship resulted for Sth via distances and the horizontal orientation angle (Ө), respectively shown in Figs. 6(c) and 7(c) respectively. Therefore, the mathematical models from fitting the Sth/Distance and Sth/theta relationships showed that such models obey the polynomial equations as shown in Eqs. (5) and (6) respectively. The appropriate Sth (Eq. (5)) and hence the suitable orientation angle (Eq. (6)) for detecting the target under investigation can be predicted.

Fig. 5. System application with increasing object distances: (a) 1m; (b) 2m; (c) 3m; (d) 5m.

Fig. 6. Variation of (a) average err; (b) ${\upsigma}$ err; (c) Sth with increasing distances in meters.}

Fig. 7. The variation of (a) average err; (b) ${\upsigma}$ err; (c) Sth with increasing the horizontal orientation angle (Ө).}

6. Conclusion

One of the resulting problems in tracking system is target detection accuracy. In the present work, the goal of detecting and tracking objects was executed efficiently with a high accuracy rate recorded a zero for the error and standard deviation for all distances. On the other hand, one can predict the true view angle to track the target under the scope. The originality of the research lies in proposing a mathematical model to predict the appropriate angle to assign the target under search.