A method and a system for classifying driver gaze using hybrid quantum-classical technique

Abstract

Disclosed herein is method and system for classifying driver gaze. The method comprises receiving at least one image related to a facial region of the driver. The method further comprises extracting, using a pre-trained Artificial Intelligence (AI) model, one or more latent vectors associated with the at least one image. Further, the method comprises fragmenting the one or more latent vectors into one or more serialized latent vector chunks based on numbers of qubits associated with a quantum computing model. Moving ahead, the method comprises determining, using the quantum computing model, one or more quantum-enhanced feature vectors by sequentially processing the one or more serialized latent vector chunks. Lastly, the method further comprises aggregating, using the pre-trained AI model, the one or more quantum-enhanced feature vectors to classify the driver gaze.

Description

Applicant’s Ref.: 2024PF02149“A METHOD AND A SYSTEM FOR CLASSIFYING DRIVER GAZE USING HYBRID QUANTUM-CLASSICAL TECHNIQUE”TECHNICAL FIELD

[0001] The present invention generally relates to the field of classical computing and quantum computing, and more particularly relates to a method and a system for classifying driver gaze using hybrid quantum-classical technique.BACKGROUND

[0002] The following description includes information that may be useful in understanding the present invention. It is not an admission that any of the information provided herein is prior art or relevant to the presently claimed invention, or that any publication specifically or implicitly referenced is prior art.

[0003] Driver gaze plays a key role in different gaze-based applications, such as driver attentiveness detection, visual distraction detection, gaze behavior understanding, and building driver assistance system.

[0004] Generally, driver in a vehicle watches front while driving the vehicle. However, in several scenarios the driver may have to watch region of interest in an inner front side of the vehicle. The region of interest may refer to a region viewed by the driver inside the vehicle. The inner front side of the vehicle may be divided into multiple regions of interest. For instance, left side mirrors of the vehicles may be assigned as seventh region of interest and right-side mirrors of the vehicles may be assigned as sixth region of interest. In other instances, the driver may have to watch middle mirror which may be assigned as eighth region of interest. Further, in other instance the driver may have to watch a display of the vehicle which may be assigned as fourth region of. Thus, there is a requirement for tracking a driver gaze and classify the driver gaze based on the region of interest viewed by the driver. Further, tracking the driver gaze may be help full to detect distraction of the driver, fatigue state of the driver, drowsiness state of the driver, thereby allowing for timely alerts and interventions to prevent accidents.Applicant’s Ref.: 2024PF02149

[0005] Conventionally, the driver gaze classification may be performed using a classical computing model. However, the classical computing model requires large number of data for training the classical computing model. Further, the classical computing model may consume time for processing the larger number of data and may not be accurate. Thus, in some conventional cases the classical computing model may be replaced with quantum computing model as the quantum computing model process the data fast, more expressive, generalize the data better, and robust against adversarial attacks. However, the quantum computing model are currently made up of very few, short-lived and noisy qubits, aptly called Noisy Intermediate-Scale Quantum (NISQ) Computers. Further, process of loading and training large datasets to the quantum computing model is difficult, as there is huge overhead of “read-in” and “read-out” time to load data from classical computing model to quantum computing model. Furthermore, as lifetime of the qubits are small, executing deep circuits are not possible, as they decohere the quantum computing model process, and the qubits may lose quantum properties to provide advantages of the quantum computing model process. Unlike classical bits in the classical computing model, two qubits in the quantum computing model process may not be added, hence size of the layers in the quantum computing model process remain equal to the input dimensions. However, the existing quantum computing model on High Performance Computing (HPC) may not process more than 40 qubits and the speed of execution is slow compared to traditional classical algorithms.

[0006] There are other techniques available in the art for classifying the driver’s gaze. However, such techniques suffer from lack of reliability, lack of accuracy and lack of efficiency. Hence, there is a requirement for accurately classifying the driver’s gaze to increase the driver’s and passengers' safety and experience while driving the vehicle.

[0007] Document US20230206108A1 describes machine learning on quantum computers, by reducing input features and then grouping the features using random sampling features, performing Principal Component Analysis (PCA). Document US20220121998A1 describes performing quantum machine learning by converting classical data-points into a quantum data-point using Quantum kernels. Further, the quantum Kernels are trained using standard Kernel methods such as Support Vector Machine (SVM), Gaussian Processes, etc. Document CN117649560A describes perform the quantum machine learning by training a Hybrid Quantum Classical Neural Network. TheApplicant’s Ref.: 2024PF02149Classical layers are trained with Stochastic Gradient Descent. The quantum layers are trained using Optimization algorithms such as Particle Swarm Optimization. Document “Quantum-Classical Hybrid Machine Learning for Image Classification (ICCAD Special Session Paper)," introduces Quanvolution and Multi-Layer Perceptron (MLP) architecture for performing image classification. The Quanvolution layer is a quantum version of convolutional Layer in Deep neural networks (DNNs). Document “Quantum Machine Learning for Image Classification” introduces Hybrid Quantum Neural Network with Quanvolutional layer to achieves state-of-the-art on Modified National Institute of Standards and Technology (MNIST) dataset. This document may receive only one image as input.

[0008] Non-patent literature relevant to the present invention includes the technical paper by Belis V et al., titled "Guided Quantum Compression for high dimensional data classification" published in journal "Machine Learning: Science and Technology", vol. 5, no. 3, 01 September 2024. The paper proposes a hybrid classical-quantum model called Guided Quantum Compression (GQC) that jointly learns data compression and classification, outperforming traditional quantum and deep learning methods on highdimensional tasks like Higgs boson detection.

[0009] Non-patent literature relevant to the present invention includes the technical paper by Belis V et al., titled "Guided Quantum Compression for Higgs Identification" published in arXiv.org, Cornell Unversity Library, 14 February 2024. The paper introduces a novel approach called Guided Quantum Compression (GQC) for Higgs boson identification. GQC combines classical dimensionality reduction with quantum classification into a single, jointly trained model, overcoming the limitations of separate preprocessing.

[0010] Therefore, there exists a need for a technique that addresses at least the above identified problems and accurately classifies the driver’s gaze while driving the vehicle.SUMMARY

[0011] The present disclosure overcomes one or more shortcomings of the prior art and provides additional advantages. Embodiments and aspects of the disclosure described in detail herein are considered a part of the claimed disclosure.Applicant’s Ref.: 2024PF02149

[0012] In one non-limiting embodiment of the present disclosure, a method for classifying driver gaze, is disclosed. The method comprises receiving at least one image related to a facial region of the driver. The method further comprises extracting, using a pretrained Artificial Intelligence (Al) model, one or more latent vectors associated with the at least one image. Further, the method comprises fragmenting the one or more latent vectors into one or more serialized latent vector chunks based on numbers of qubits associated with a quantum computing model. Moving ahead, the method comprises determining, using the quantum computing model, one or more quantum- enhanced feature vectors by sequentially processing the one or more serialized latent vector chunks. Lastly, the method further comprises aggregating, using the pre-trained Al model, the one or more quantum-enhanced feature vectors to classify the driver gaze. Thus, a hybrid quantum classical technique is used to classify driver gaze.

[0013] In another non-limiting embodiment of the present disclosure, wherein for aggregating the one or more quantum-enhanced feature vectors to classify the driver gaze, the method comprises determining a centroid feature vector for each of the one or more quantum-enhanced feature vectors. The method further comprises determining a variance feature vector for each centroid feature vector associated the one or more quantum-enhanced feature vectors. Lastly, the method further comprises aggregating the one or more quantum-enhanced feature vectors by applying a contrastive loss function on each variance feature vector to classify the driver gaze into at least one predefined zone inside vehicle.

[0014] In yet another non-limiting embodiment of the present disclosure, wherein for applying the contrastive loss function, the method comprises modifying the contrastive loss function by replacing a temperature corresponding to the contrastive loss function with an inverse of the variance feature vector associated to one or more quantum-enhanced feature vectors, wherein the temperature is associated with qubits of a quantum computing model.

[0015] In yet another non-limiting embodiment of the present disclosure, wherein for fragmenting the one or more latent vectors into one or more serialized latent vector chunks, the method comprises fragmenting the one or more latent vectors into one or more latent vector chunks. The method further comprises appending a position to each of the one or more latent vector chunks to form the one or more serialized latent vectorApplicant’s Ref.: 2024PF02149 chunks. Moving ahead, the method further comprises feeding the one or more serialized latent vector chunks sequentially to the quantum computing model, based on the appended position.

[0016] In yet another non-limiting embodiment of the present disclosure, wherein for appending the position to form the one or more serialized latent vector chunks, the method further comprises sequencing the one or more serialized latent vector chunks. The method further comprises performing an angle encoding on each of the one or more serialized latent vector chunks to process the position of each of the one or more serialized latent vector chunks, upon sequencing the one or more serialized latent vector chunks. Moving ahead, the method further comprises determining the one or more quantum-enhanced feature vectors for the one or more serialized latent vector chunks based on the processed position.

[0017] The present disclosure proposes to use both classical computing model and quantum computing model for classifying driver gaze by receiving a facial region of the driver as input. Thus, the combination of both the classical computing model and quantum computing model help is classifying the driver gaze more accurately and efficiently thereby increasing the driver’s and passengers' safety and experience while driving a vehicle. The contrastive loss function helps in training the hybrid quantum-classical model faster. Thereby, increasing the processing speed of the hybrid quantum-classical model. Further, replacing the temperature with the inverse of variance feature vector may eliminate poor latent feature-vectors contributing to overall loss. Further, this greatly enhances the training process. Angle encoding helps the quantum computing model to extra the one or more quantum-enhanced feature vectors at the cost of just an extra qubit. Thereby, the present disclosure provides higher accuracy on training with the one or more quantum-enhanced feature vectors.

[0018] In one non-limiting embodiment of the present disclosure, a system for classifying driver gaze, is disclosed. The system comprises a memory, processor coupled to the memory and a quantum processor coupled to the memory. The processor receives at least one image related to a facial region of the driver. The processor further extracts, using the pre-trained Artificial Intelligence (Al) model, one or more latent vectors associated with the at least one image. The processor further fragments, the one or more latent vectors into one or more serialized latent vector chunks based on numbers ofApplicant’s Ref.: 2024PF02149 qubits associated with the quantum computing model. Moving ahead, the quantum processor determines using the quantum computing model, one or more quantum- enhanced feature vectors by sequentially processing the one or more serialized latent vector chunks. Lastly, the processor aggregates, using the pre-trained Al model, the one or more quantum-enhanced feature vectors to classify the driver gaze.

[0019] In yet another embodiment of the present disclosure, wherein to aggregate the one or more quantum-enhanced feature vectors to classify the driver gaze, the processor determines a centroid feature vector for each of the one or more quantum-enhanced feature vectors. Further, the processor determines a variance feature vector for each centroid feature vector associated the one or more quantum-enhanced feature vectors. Moving ahead, the processor aggregates the one or more quantum-enhanced feature vectors by applying a contrastive loss function on each variance feature vector to classify the driver gaze into at least one predefined zone inside vehicle. Thus, helps in classifying the driver gaze more accurately.

[0020] In yet another embodiment of the present disclosure, wherein to apply the contrastive loss function, the processor modifies the contrastive loss function by replacing a temperature corresponding to the contrastive loss function with an inverse of the variance feature vector associated to one or more quantum-enhanced feature vectors, wherein the temperature is associated with qubits of a quantum computing model.

[0021] In yet another embodiment of the present disclosure, wherein to fragment the one or more latent vectors into one or more serialized latent vector chunks, the processor fragments the one or more latent vectors into one or more latent vector chunks. The processor further appends a position to each of the one or more latent vector chunks to form the one or more serialized latent vector chunks.

[0022] In yet another embodiment of the present disclosure, to append the position to form the one or more serialized latent vector chunks, the processor sequences the one or more serialized latent vector chunks. The processor further performs an angle encoding on each of the one or more serialized latent vector chunks to process the position of each of the one or more serialized latent vector chunks, upon sequencing the one or more serialized latent vector chunks. The processor then determines the one or moreApplicant’s Ref.: 2024PF02149 quantum-enhanced feature vectors for the one or more serialized latent vector chunks based on the processed position.

[0023] The foregoing summary is illustrative only and is not intended to be in any way limiting. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features will become apparent by reference to the drawings and the following detailed description.BRIEF DESCRIPTION OF DRAWINGS

[0024] The features, nature, and advantages of the present disclosure will become more apparent from the detailed description set forth below when taken in conjunction with the drawings in which like reference characters identify correspondingly throughout. Some embodiments of system and / or methods in accordance with embodiments of the present subject matter are now described, by way of example only, and with reference to the accompanying Figs., in which:

[0025] FIG. 1 depicts an exemplary environment classifying driver gaze using hybrid quantum-classical technique, in accordance with embodiments of the present disclosure.

[0026] FIG. 2 depicts an exemplary block diagram illustrating a system for classifying driver gaze using hybrid quantum-classical technique, in accordance with embodiments of the present disclosure.

[0027] FIG. 3 illustrates a flow diagram for classifying driver gaze using hybrid quantum- classical technique, in accordance with embodiments of the present disclosure.

[0028] FIG. 4a illustrates a block diagram for extracting latent vectors, in accordance with embodiments of the present disclosure.

[0029] FIG. 4b illustrates a block diagram for appending position to fragmented chunks, in accordance with embodiments of the present disclosure.

[0030] FIG. 4c illustrates a process for feeding an angle encoded fragmented chunks to quantum computing model, in accordance with embodiments of the present disclosure.Applicant’s Ref.: 2024PF02149

[0031] Fig. 5 represents a flowchart of an exemplary method for classifying driver gaze using hybrid quantum-classical technique, in accordance with embodiments of the present disclosure.

[0032] It should be appreciated by those skilled in the art that any block diagrams herein represent conceptual views of illustrative systems embodying the principles of the present subject matter. Similarly, it will be appreciated that any flow charts, flow diagrams, state transition diagrams, pseudo code, and the like represent various processes which may be substantially represented in a computer readable medium and executed by a computer or processor, whether or not such computer or processor is explicitly shown.DETAILED DESCRIPTION

[0033] The foregoing has broadly outlined the features and technical advantages of the present disclosure in order that the detailed description of the disclosure that follows may be better understood. It should be appreciated by those skilled in the art that the conception and specific embodiment disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure.

[0034] The novel features which are believed to be characteristic of the disclosure, both as to its organization and method of operation, together with further objects and advantages will be better understood from the following description when considered in connection with the accompanying figures. It is to be expressly understood, however, that each of the figures is provided for the purpose of illustration and description only and is not intended as a definition of the limits of the present disclosure.

[0035] As described earlier, the classical computing model requires large number of data for training the classical computing model for classifying the driver gaze. Also, it is difficult to load and train large datasets to the quantum computing model, as there is huge overhead of “read-in” and “read-out” time to load data from classical devices to quantum hardware and as lifetime of the qubits are small. Also, the execution speed of existing quantum simulators is slow as compared to the classical models running on a classical computer. Thus, there is a requirement for efficiently and accuratelyApplicant’s Ref.: 2024PF02149 classifying the driver’s gaze to increase the driver’s and passengers' safety and experience while driving the vehicle.

[0036] The present disclosure provides a method and a system for classifying driver gaze using hybrid quantum-classical technique. In particular, the present disclosure proposes a method with the combination of both classical and quantum computing model to classify the driver gaze. In the present disclosure, the classical processor may extract features from the images received as input. Further, the features are fragmented into serialized chunks. Then, the quantum computing model determines quantum enhanced features from the serialized chunks. Finally, the quantum enhanced features are aggregated to classify the driver’s gaze. Thus, the combination of both classical and quantum computing models provides accurate classification of the driver’s gaze with improved efficiency. Thereby increasing the driver’s and passengers' safety and experience while driving the vehicle.

[0037] FIG. 1 exemplary environment 100 for classifying driver gaze, in accordance with embodiments of the present disclosure. The exemplary environment 100 particularly depicts the vehicle 102 that may incorporate a system 104. The environment 100 is exemplified for a scenario when a driver of the vehicle 102 is driving and may view at least one predefined zone inside the vehicle 102. In an example without limitation, the at least one predefined zone may refer to certain areas inside the vehicle 102 as shown in Fig. 3 (i.e., vehicle zone labels 310). The at least one predefined zone may be viewed by the driver while driving the vehicle 102. For instance, the driver may view left side mirror or a right-side mirror. In other instance, the driver may view a front-end display of the vehicle 102. In an example without limitation, the at least one predefined zone may be divided into 9 zones as shown in Fig. 3 (i.e., vehicle zone labels 310). A person of ordinary skill will appreciate that the areas on the inside front of the vehicle may be divided into any suitable number of zones as per the requirement of the present disclosure.

[0038] In some implementations, the system 104 may comprise a processor 106, a memory 108, an Artificial Intelligence (Al) model 110 and a quantum processor 112. In some implementations, the system 104 may include other components (not shown in fig. 1) to implement desired functions of the system 104. In an exemplary embodiment, the Al model 110 may be pretrained. The quantum processor 112 may further comprise aApplicant’s Ref.: 2024PF02149 quantum computing model 112a. The processor 106 may classify the driver gaze into the at least one predefined zone inside vehicle 102 by receiving at least one image related a facial region of the driver. The system 104 may classify the driver gaze by aggregating the quantum-enhanced feature vector form the at least one image.

[0039] In an exemplary embodiment, the processor 106 may receive the at least one image related to the facial region of the driver. In an exemplary embodiment, the facial region may be but not limited thereto to eyes of the driver, a face of the driver and the like. In an exemplary embodiment, the at least one image may be received from one or more datasets. In an exemplary embodiment, the at least one image may be pre-processed by the processor 106. Upon receiving the at least one image as input, the processor 106 may using the Al model 110 to extract one or more latent vector associated with the at least one image. The Al model may be pre-trained for extracting the one or more latent vectors. Then, the processor 106 may fragment the one or more latent vectors to form one or more serialized latent vector chunks based on numbers of qubits associated with a quantum computing model. In an exemplary embodiment, the one or more latent vectors may be fragmented and appended with a position to form the one or more serialized latent vector chunks. Then, the quantum processor 112 may determine one or more quantum-enhanced feature vectors using the quantum computing model 112a. After determining the one or more quantum-enhanced feature vectors, the processor 106 may use the pre-trained Al model 110 for aggregating the one or more quantum- enhanced feature vectors by applying a contrastive loss function to classify the driver gaze into the at least one predefined zone inside the vehicle 102. A detailed explanation of the system 104 is provided in the forthcoming paragraphs in conjunction with FIG.s 2, 3-5.

[0040] Fig. 2 depicts an exemplary block diagram illustrating a system 200 (which is system 104 of Fig. 1) for classifying the driver gaze into the at least one predefined zone inside vehicle 102 (of Fig. 1), in accordance with embodiments of the present disclosure. In an exemplary embodiment, the system 200 may be a classical computer. In some implementations, the system 200 may further comprise one or more sensors 202, a processor 204, a memory 206, an Artificial Intelligence (Al) model 208 and a quantum processor 210. The memory 206 may further comprise input data 206a and other data 206b. The quantum processor 210 may further comprise a quantum computing modelApplicant’s Ref.: 2024PF02149210a. In an exemplary embodiment, the Al model 208 and the quantum processor 210 may be coupled with the processor 204. In another exemplary embodiment, the Al model 208 and the quantum processor 210 may be separated from the processor 204.

[0041] In one implementation, one or more sensors 202 may be installed on the vehicle 102. In an exemplary embodiment, while the vehicle 102 is moving, the one or more sensors 202 may continuously capture at least one image related to facial region of the driver in the vehicle 102. In an example without limitation, the one or more images may be a face of the driver, the left eye of the driver, a right eye of the driver. The non-limiting examples of the one or more sensors 202 may be a camera, a forward-facing camera and the like. The one or more sensors 202 may provide the at least one image to the processor 204 of the system 200 for further processing.

[0042] In one implementation, the processor 204 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuitries, and / or any devices that manipulate signals based on operational instructions. Among other capabilities, the processor 204 may be configured to fetch and execute computer-readable instructions and other information stored in the memory 206. In an exemplary embodiment, the processor 204 may be used for execution instruction of a classical computer. In another exemplary embodiment, the processor 204 may be a general processor used for executing the instructions.

[0043] In an exemplary embodiment, the memory 206 may include one or more non-transitory computer-readable storage media, such as RAM, ROM, EEPROM, EPROM, one or more memory devices, flash memory devices, etc., and combinations thereof. In an exemplary embodiment, the input data 206a may be stored within the memory 206 in the form of various data structures. In a non-limiting example, input data 206a refers as plurality of inputs relating to the at least one image received as input from the one or more sensors 202 and the like. The memory 206 may also store other data 206b such as temporary data and temporary files, generated by the processor 204 or other any other parts of the system 200 including the Al model 208 and the quantum processor 210 for performing the various functions of the present invention.Applicant’s Ref.: 2024PF02149

[0044] In an exemplary embodiment, the Al model 208 may be a pretrained for performing particular task use a set of technologies. A person of ordinary skill will appreciate that the Al model 208 may use the input data to mimic human cognitive functions like learning, problem-solving, and reasoning. As the Al model 208 may be pre-trained to perform particular task, hereafter the Al model 208 is referred as pre-trained Al model 208 In an exemplary embodiment, the pre-trained Al model 208 may reside inside the processor 204. In another exemplary embodiment, the Al model 208 may reside outside the processor 204. Further, in the classical computer the data may be store in bits.

[0045] In an exemplary embodiment, the quantum processor 210 may refer to a quantum computer that uses quantum mechanics to perform computation on input data to solve problems faster than classical computers. A person of ordinary skill will appreciate that the quantum computer may store data in qubits. Further, the quantum computer is maintained with low temperature to cool the system that in turn helps in minimizing vibrations of the qubits associated with the quantum computer and reducing the impact of thermal noise and enhancing the stability of qubits. The quantum computing model 210a may reside within the quantum processor for performing computation on the qubits associated with quantum processor 210 to solve problems. In an exemplary embodiment, the quantum processor 210 may be coupled with the processor 204. In another exemplary embodiment, the quantum processor 210 may separately compute to solve problems.

[0046] In an exemplary embodiment, the processor 204 may receive the at least one image related to the facial region of the driver in the vehicle 102, from the one or more sensor 202. The facial region may not be limited to the face of the driver, the left eye of the driver and the right eye of the driver. In another exemplary embodiment, the processor 204 may receive the at least one pre-processed image from one or more datasets.

[0047] Referring to Fig. 3, facial region 302 may include but is not limited to a face crop 302a, a left eye 302b, a right eye 302c. Then, the captured face crop 302a, the left eye 302b, the right eye 302c may be provided to an Al model 304 for further processing.

[0048] Referring back to Fig. 2, upon receiving the at least one image related to the facial region of the driver in the vehicle 102, the processor 204 may extract the one or more latent vectors associated with the at least one image using the pre-trained Al model 208.Applicant’s Ref.: 2024PF02149A person of ordinary skill will appreciate that the latent vectors refer to an intermediate representation of data that are often used to obtain the essence of the data. Broadly, the latent vector may be referred to as hidden features with respect to the obtained data, that may give the essence of the data. In an embodiment, the processor 204 may extract the one or more latent vectors from the pre-processed at least one image. In an example without limitation, the processor 204 may extract the one or more latent vectors from the facial region of the driver. Thereby, the pre-trained Al model 208 helps in scaling the quantum processor 210 with large datasets using pre-trained convolution layer, as the quantum processor 210 may be limited with the size and number of qubits. Thus, larger number of input images may be used for processing. In a non-limiting embodiment, the pre-trained Al model 208 may be a Visual Geometry Group (VGG), such as a VGG 16.

[0049] Referring back to fig. 3, after receiving the facial region 302, the pre-trained Al model 304 may extract the one or more latent vectors from the facial region. In an example without limitation, the pre-trained Al model 304 may extract the one or more latent vectors form each of the face crop 302a, the left eye crop 302b, the right eye crop 302c.

[0050] Referring to fig. 4a, a pre-trained Al model 402 may extract one or more latent vectors 404 form each of the at least one image received as input. In an example without limitation, the pre-trained Al model may extract latent vectors 404 such as vl, v2, v3, v4 and the like from the face crop 302a (as shown in Fig. 3).

[0051] Referring back to Fig. 2 after extracting the one or more latent vectors, the processor 204 may fragment the one or more latent vectors into the one or more serialized latent vector chunks based on the numbers of qubits associated with the quantum computing model 210a. In an exemplary embodiment, the processor 204 may fragment the one or more latent vectors into one or more into one or more latent vector chunks. The one or more latent vector may be fragmented into the one or more latent vector chunks based on the number of qubits associated with the quantum computing model 210a. In an example without limitation, if the number of qubits available to encode the one or more latent vectors are 8. Then, the number of fragmented one or more chunks may be multiple of 8. In an exemplary embodiment, the processor 204 may equally fragment the one or more latent vector into non-overlapping one or more latent vector chunks (asApplicant’s Ref.: 2024PF02149 shown in Fig. 3). Further, the processor 204 may append a position to each of the one or more latent vectors chunks to form the one or more serialized latent vector chunks.

[0052] Then, the processor 204 may append the position for each of the one or more latent vector chunks, to form the one or more serialized latent vectors. Then, the processor 204 may perform an angle encoding on each of the one or more serialized latent vector chunks to process the position of each of the one or more serialized latent vector chunks. The angle encoding transforms classical data into quantum states by representing each feature as an angle on the Bloch sphere, enabling efficient data handling in the quantum computing model. Further, the processor 204 may feed the one or more serialized latent vector chunks to the quantum processor 210, upon performing the angle encoding. The one or more serialized latent vector chunks may be sequentially fed based on the appended position to the quantum processor 210, for processing the one or more serialized latent vector chunks using the quantum computing model 210a.

[0053] Referring to Fig. 4b, each of the one or more latent vectors 404 may be fragmented to one or more latent vector chunks 406. In an example without limitation, the latent vector vl may be fragmented into one or more latent vector chunks 406 cl, c2, cn and the like. However, fragmenting the latent vector into the one or more latent vector chunks may not be limited to fragmenting the latent vector vl to the one or more latent vector chunks 406 cl, c2, cn. Then, each of the one or more latent vector chunks 406 may be appended with the position for sequentially processing the one or more latent vector chunks 406. For instance, each of the one or more latent vector chunks 406 cl, c2, cn may be appended with positions [0], [1], [n] respectively.

[0054] Referring to Fig. 4c, upon appending the position to each of the one or more latent vector chunks 406, an angle encoding 410 may be performed on each of the one or more latent vector chunks appended with position 408. In an exemplary embodiment, each of the one or more latent vector chunks may be passed via qubits associated with the quantum processor 412 for processing the position associated with each of the one or more latent vector chunks 408. Then, the angle encoded one or more latent vector chunks may be feed to a quantum processor 412 for further processing. The processor 204 may use classical convolutional layers for extracting the one or more latent vectors with overlapping. However, to reduce the read -in for the quantum processor 210, non-Applicant’s Ref.: 2024PF02149 overlapping windows of the one or more latent vectors may be utilized, making the training of neural -networks difficult. Thus, to overhead this, the angle-encoding is used, that helps the quantum processor 210 with extra information at the cost of just an extra qubit. Thereby leading to higher accuracy on training with the features.

[0055] Referring back to Fig. 2, after feeding the angle encoded one or more latent vector chunks 306, the quantum processor 308 may determine the one or more quantum- enhanced feature vectors by sequentially processing the one or more serialized latent vector chunks using the quantum computing model 210a. In a non-limiting example, the quantum processor 308 may determine the one or more quantum-enhanced feature vectors by sequentially processing the one or more serialized latent vector chunks based on their respective appended positions.

[0056] In an exemplary embodiment, the one or more quantum-enhanced feature vectors may refer to measurements associated with qubits corresponding to the quantum processor 210. In the present disclosure, the one or more quantum -enhanced feature vectors may refer to the measurements associated with each of the one or more latent vector chunks.

[0057] Referring back to Fig. 3, the quantum computing model 308 may receive the serialized one or more latent vector chucks 306 to extract the one or more quantum-enhanced feature vectors, as indicated above, each of the one or more latent vector chucks 306 may be appended with the position for sequentially processing the one or more latent vector chucks 306 by the quantum computing model 308 for extracting the one or more quantum-enhanced feature vectors. Since the one or more latent vector chucks 306 are serialized based on their appended positions before these are fed to the quantum processor 210, the processing efficiency of the quantum processor 210 in determining the one or more quantum-enhanced feature vectors, is improved.

[0058] Referring back to Fig. 4c, the quantum processor 412 may extract the one or more quantum-enhanced feature vectors using the quantum computing model 412a, upon appending the position by the processor 204 to each of the one or more latent vector chunks, for further processing.

[0059] Referring back to Fig. 2, in an exemplary embodiment, upon determining the one or more quantum-enhanced feature vector, the processor 204 may aggregate the one orApplicant’s Ref.: 2024PF02149 more quantum-enhanced feature vectors using the pre-trained Al model 208 to classify the driver gaze. In an exemplary embodiment, upon determining the one or more quantum-enhanced feature vectors, the processor 204 may determine a centroid feature vector for each of the one or more quantum-enhanced feature vectors. Further, the processor may determine a variance feature vector for each of the centroid feature vector associated with the one or more quantum-enhanced feature vectors. Then, the processor 204 may aggregate the one or more quantum-enhanced feature vectors by applying a contrastive loss function on each variance feature vector to classify the driver gaze into at least one predefined zone inside vehicle. In an exemplary embodiment, InfoNCE-Contrastive loss function for aggregating the variance feature vectors.

[0060] In the present disclosure, the contrastive loss function may be applied using the Al model 208, by modifying the contrastive loss function. The contrastive loss function may be modified by replacing a temperature corresponding to the contrastive loss function with an inverse of the variance feature vector associated to one or more quantum-enhanced feature vectors. In an exemplary embodiment, the temperature may be associated with qubits of a quantum computing model. In a non-limiting example, if the one or more quantum-enhanced feature vectors are good then the temperature may be reduced, and the impact will be high. In another non-limiting example, if the one or more quantum-enhanced feature vectors are bad then the temperature may be increased, and the impact will be low. The contrastive loss function may increase the processing speed of the Al model 208. Thereby, the increasing performance of the system 104. Further, modifying the temperature with the inverse of the variance feature vector may eliminate poor one or more latent vector chunks, that may lead to overall loss of the performance of the system 104 and thereby greatly enhance the processing speed of the system 200. Further, the contrastive loss function may reduce the overall wall-time for processing the data for classifying the driver gaze. Also, Fl score of the system may be increased due to the contrastive loss function. Fl score may refer to harmonic score of a model. Further, the processor 204 may perform fully supervised learning using cross entropy loss along with the InfoNCE-Contrastive loss function. In an exemplary embodiment, the InfoNCE-Contrastive loss function may as per below equation (1).Applicant’s Ref.: 2024PF02149Wherein, z(- : Quantum featuresPt : Positive samples cii : Positive and Negative samplesT : Temperature which we take as inverse of variance from the Quantum measurement.

[0061] The processor 204 may process the quantum-enhanced feature vectors using a fully connected layer for classifying the driver gaze into at least one predefined zone inside the vehicle 102. In an exemplary embodiment, the final layer of the fully connected layer may be SoftMax layer that may be used for classifying the driver gaze, the one or more quantum-enhanced feature vectors may be aggregated to classify the driver gaze into at least one predefined zone 310 inside the vehicle 102 (as shown in Fig. 3).

[0062] FIG. 5 represents flowchart of an exemplary method for classifying the driver gaze, in accordance with embodiments of the present disclosure. The order in which the method 500 is described is not intended to be construed as a limitation, and any number of the described method blocks may be combined in any order to implement the method. Additionally, individual blocks may be deleted from the methods without departing from the spirit and scope of the subject matter described. Furthermore, the method can be implemented in any suitable hardware, software, firmware, or combination thereof. However, for ease of explanation, in the embodiments described below, the method 500 may be implemented by the respective components and / or by the processor 204 using the the Al model 208 and the quantum processor 210 using the quantum computing model 210a of FIG. 2.

[0063] At step 502, the method may include receiving the at least one image related to the facial region of the driver. The at least one image may be received as input by the system 104. In one implementation, the processor 204 may receive the at least one image related to facial region.

[0064] At step 504, the method may include extracting the one or more latent vector associated with the at least one image. In one implementation, the processor 204 may extract the latent vectors may be extracted using the pretrained Al model 110.Applicant’s Ref.: 2024PF02149

[0065] At step 506, the method may include fragmenting the one or more latent vectors into one or more serialized latent vector chunks based on numbers of qubits associated with a quantum computing model. In one implementation, the processor 204 may fragment the one or more latent vectors into one or more serialized latent vector chunks. Then, the processor 106 may append the position to each of the one or more latent vector chunks to form the one or more serialized latent vector chunks.

[0066] At step 508, the method may include determining the one or more quantum-enhanced feature vectors. In one implementation, the quantum processor 210 may use the quantum computing model 210a for determining the one or more quantum-enhanced feature vectors. The quantum processor 112 may determine the one or more quantum- enhanced feature vectors by sequentially processing the one or more serialized latent vector chunks.

[0067] At step 510, the method may include aggregating the one or more quantum-enhanced feature vectors to classify the driver gaze. In one implementation, the processor 204 may use the Al model 208 for aggregating the one or more quantum -enhanced feature vectors. In an exemplary embodiment, the processor 106 may aggregate the the one or more quantum-enhanced feature vectors by applying the contrastive loss function on the one or more quantum-enhanced feature vectors to classify the driver gaze. In the present disclosure the contrastive loss function may be modified by replacing the temperature with the inverse of the variance to aggregate the one or more quantum- enhanced feature, to classify the driver gaze more accurately gaze into at least one predefined zone inside vehicle 102.

[0068] The order in which the method 500 is described is not intended to be construed as a limitation, and any number of the described method blocks may be combined in any order to implement the method. Additionally, individual blocks may be deleted from the methods without departing from the spirit and scope of the subject matter described.

[0069] The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functionalApplicant’s Ref.: 2024PF02149 building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed.

[0070] Advantages of the present disclosure:• Improved reliability, accuracy, and efficiency in classifying the driver’s gaze.• The processor using the Al model helps in scaling the quantum processor with large datasets by adding pre-trained convolution layer, as the quantum processor may be limited with the size and number of qubits. Thus, larger number of input images may be used for processing.• To alleviate the huge overhead of read-in and read-out of data to the quantum processors, the angle-encoding is used, that helps the quantum processor to extra information at the cost of just an extra qubit. Thereby leading to higher accuracy on training with the features.• The contrastive loss function may increase the processing speed of the Al model. Thereby, the increasing performance of the system. Further, modifying the temperature with the inverse of the variance feature vector may eliminate poor latent vector chunks, that may lead to overall loss of the performance of the system 104 and thereby greatly enhance the processor the system. Further, the contrastive loss function may reduce the overall wall-time for processing the data for classifying the driver gaze.

[0071] Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope and spirit of the disclosed embodiments. Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer- readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., are non-transitory. Examples include random access memory (RAM), read-onlyApplicant’s Ref.: 2024PF02149 memory (ROM), volatile memory, non-volatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.

[0072] REFERENCE NUMERALS

Claims

Applicant’s Ref.: 2024PF02149We Claim:

1. A method for classifying driver gaze, comprising: receiving at least one image related to a facial region of the driver; extracting, using a pre-trained Artificial Intelligence (Al) model (110), one or more latent vectors associated with the at least one image; fragmenting the one or more latent vectors into one or more serialized latent vector chunks based on numbers of qubits associated with a quantum computing model (112a); determining, using the quantum computing model (112a), one or more quantum- enhanced feature vectors by sequentially processing the one or more serialized latent vector chunks; and aggregating, using the pre-trained Al model (110), the one or more quantum-enhanced feature vectors to classify the driver gaze.

2. The method as claimed in claim 1, wherein aggregating the one or more quantum-enhanced feature vectors to classify the driver gaze, comprising: determining a centroid feature vector for each of the one or more quantum-enhanced feature vectors; determining a variance feature vector for each centroid feature vector associated with the one or more quantum-enhanced feature vectors; and aggregating the one or more quantum-enhanced feature vectors by applying a contrastive loss function on each variance feature vector to classify the driver gaze into at least one predefined zone inside vehicle (102);3. The method as claimed in claim 2, wherein applying the contrastive loss function further comprising: modifying the contrastive loss function by replacing a temperature corresponding to the contrastive loss function with an inverse of the variance feature vector associated to one or more quantum-enhanced feature vectors, wherein the temperature is associated with qubits of a quantum computing model (112a).

4. The method as claimed in claim 1, wherein fragmenting the one or more latent vectors into one or more serialized latent vector chunks, comprising: fragmenting the one or more latent vectors into one or more latent vector chunks;Applicant’s Ref.: 2024PF02149 appending a position to each of the one or more latent vector chunks to form the one or more serialized latent vector chunks; and feeding the one or more serialized latent vector chunks sequentially to the quantum computing model (112a), based on the appended position.

5. The method as claimed in claim 4, wherein appending the position to form the one or more serialized latent vector chunks further comprising: sequencing the one or more serialized latent vector chunks; performing an angle encoding on each of the one or more serialized latent vector chunks to process the position of each of the one or more serialized latent vector chunks, upon sequencing the one or more serialized latent vector chunks; and determining the one or more quantum-enhanced feature vectors for the one or more serialized latent vector chunks based on the processed position.

6. A system (104) for classifying driver gaze, comprises: a memory (108); processor (106) coupled to the memory (108); and a quantum processor (112) coupled to the memory (108), the processor (106) and the quantum processor (112) are configured to: receive at least one image related to a facial region of the driver; extract using a pre-trained Artificial Intelligence (Al) model (110), one or more latent vectors associated with the at least one image; fragment the one or more latent vectors into one or more serialized latent vector chunks based on numbers of qubits associated with a quantum computing model (112a); determine using the quantum computing model (112a), one or more quantum- enhanced feature vectors by sequentially processing the one or more serialized latent vector chunks; and aggregate using the pre-trained Al model (110), the one or more quantum- enhanced feature vectors to classify the driver gaze.

7. The system (104) as claimed in claim 6, wherein to aggregate the one or more quantum- enhanced feature vectors to classify the driver gaze, the processor (106) is configured to: determine a centroid feature vector for each of the one or more quantum-enhanced feature vectors;Applicant’s Ref.: 2024PF02149 determine a variance feature vector for each centroid feature vector associated the one or more quantum-enhanced feature vectors; and aggregate the one or more quantum-enhanced feature vectors by applying a contrastive loss function on each variance feature vector to classify the driver gaze into at least one predefined zone inside vehicle (102);8. The system (104) as claimed in claim 7, wherein to apply the contrastive loss function, the processor (106) is configured to: modify the contrastive loss function by replacing a temperature corresponding to the contrastive loss function with an inverse of the variance feature vector associated to one or more quantum-enhanced feature vectors, wherein the temperature is associated with qubits of a quantum computing model (112a).

9. The system (104) as claimed in claim 6, wherein to fragment the one or more latent vectors into one or more serialized latent vector chunks, the processor (106) is configured to: fragment the one or more latent vectors into one or more latent vector chunks; append a position to each of the one or more latent vector chunks to form the one or more serialized latent vector chunks; and feed the one or more serialized latent vector chunks sequentially to the quantum computing model (112a), based on the appended position.

10. The system (104) as claimed in claim 9, wherein to append the position to form the one or more serialized latent vector chunks, the processor (106) is configured: sequence the one or more serialized latent vector chunks; perform an angle encoding on each of the one or more serialized latent vector chunks to process the position of each of the one or more serialized latent vector chunks, upon sequencing the one or more serialized latent vector chunks; and determine the one or more quantum-enhanced feature vectors for the one or more serialized latent vector chunks based on the processed position.

Images