Asynchronous pixel matrix for an image with automatic exposure (AE) and high dynamic range (HDR) in a single capture

Abstract

The present invention relates to a system and a method for capturing images with simultaneous automatic exposure (AE) and high dynamic range (HDR) in a single shot, simultaneously performing AE and HDR during the capture of an image. For each photointegration period established to produce the image, the circuitry of a pixel matrix automatically adjusts its response to a mean lighting at the time and adapts a radiation map of the scene within an available pixel signal range.

Description

[0001]ASYNCHRONOUS PIXEL ARRAY FOR AN IMAGE WITH AUTOMATIC EXPOSURE (AE) AND HIGH DYNAMIC RANGE (HDR) IN A SINGLE SHOT DESCRIPTION OBJECT OF THE INVENTION The object of the present invention is an asynchronous pixel array for an image with simultaneous automatic exposure (AE) and high dynamic range (HDR) in a single shot, based on a detection strategy that simultaneously performs AE and HDR during image capture. BACKGROUND OF THE INVENTION Ideally, image sensors are supposed to provide a faithful representation of the scene, regardless of the lighting conditions at the time of capture.This is a significant challenge for several reasons, two of which are particularly relevant to the present invention: (1) the sensor must adapt its response to the ambient illumination, which is not known a priori; and (2) the sensor must deal with the dynamic range of real-world scenes, which can be extremely high and is also not known in advance. Figure 1 illustrates how these points (1) and (2) are typically addressed in standard CMOS image sensors. First, the sensor's response, which is approximately linear and typically ranges from 60 to 70 dB, is shifted to the appropriate range according to some particular aspect of the scene (e.g., in the simplest case, according to the average illumination). Auto-exposure (AE) techniques are typically employed for this purpose. Second, if the scene to be imaged requires a higher dynamic range than the sensor—e.g.,, an exterior window in a dimly lit room –, what is known as exposure bracketing is applied: multiple images are captured at different exposure times and then combined to produce a high dynamic range (HDR) image. The above approaches work well for static scenes. However, in scenes characterized by abrupt light changes or movement, AE algorithms may not converge appropriately, resulting in inappropriate exposure settings and / or flickering (fluctuations in image brightness) while exposure bracketing leads to motion artifacts. Furthermore, both approaches prolong the image formation process while trying to establish the optimal exposure and while rendering the HDR image.This can lead to the loss of crucial scene information when sudden variations occur, for example, in surveillance and security applications or in robot navigation using computer vision. AE (or automatic exposure) refers to the automatic adjustment of exposure parameters—usually their duration—to produce an optimal image. There is no exact definition of the term "optimal" in this context. This is why, in most commercially available devices that include image capture, for example, mobile phones, users are allowed to guide the AE algorithms according to personal aesthetic criteria.These algorithms rely on a variety of metrics and statistical tools, such as luminance histogram, image gradient, image entropy, or object detection accuracy, to determine an appropriate exposure time through an iterative process typically based on a feedback loop. This loop requires some time to converge, which can be critical in scenarios where decisions must be made as quickly as possible, e.g., in drone navigation or autonomous vehicles. Furthermore, constant changes in a dynamic scene can further delay convergence, or even make it impossible, thus exacerbating the problem. Properly centering the sensor's response using AE techniques is not sufficient in scenarios characterized by a range of illuminations that extends well beyond the sensor's dynamic range.In such scenarios, the most common approach to avoid overexposed or underexposed regions in the resulting image is the exposure bracketing technique mentioned above. The main advantage of this approach is that it exploits the linear response of standard CMOS image sensors; it is therefore an inexpensive solution that requires no additional hardware on the sensor. However, several consecutive captures are required to cover the scene's illumination range. These captures must subsequently be merged to produce the HDR representation. The lengthy process required to complete this image fusion generates a high computational load and entails high memory consumption, not to mention leading to poor encoding of dynamic scenes. To mitigate these issues, HDR-specific pixel implementations have been proposed.In general, these embodiments require specific AE techniques to establish appropriate exposure settings, although there are exceptions. For example, sensors based on self-reset and saturation event counting accumulate information during the photointegration interval to subsequently reconstruct the corresponding HDR image, regardless of the lighting conditions; their fundamental limitations in terms of dynamic range are noise, the resolution of the counter, and the refresh rate required by the target application. HDR sensors based on logarithmic response pixels are another approach that does not require AE. In this case, there is no integration period because the light intensity is sampled continuously.Also within this approach with logarithmic response pixels, photovoltaic pixels introduce a performance improvement. In contrast, the present invention presents a method that fuses AE and HDR during image capture, overcoming some of the limitations of previous methods. In contrast to exposure bracketing, the proposed method requires a single capture to quickly provide a representation of the scene, ready for subsequent analysis. Compared to self-resetting sensors, in our case, circuitry for counting the number of saturations and rendering the corresponding image is not required—the operating principle is completely different—but a linear pixel response has been maintained as a fundamental advantage.Finally, although logarithmic sensors can theoretically encode any irradiance map, there is no inherent adaptation to the lighting conditions at the time of capture. This can lead to poor use of the available signal range and consequent generation of a noisy image; an extra tone mapping stage might even be required to allow for visual processing. In contrast, in the proposed approach, implicit adaptation to the mean illumination automatically centers the sensor response; and HDR images are achieved around this reference illumination. SUMMARY OF THE INVENTION The present invention is based on a detection strategy that simultaneously performs high dynamic range (HDR) capture and automatic exposure (AE).At each photointegration period established to obtain an image, the pixel array circuitry automatically adjusts its response to the average illumination at each moment and simultaneously adjusts the scene's light intensity map within the signal range available at the pixel. The sensor's response is linear for low light intensity levels. The terms "low" and "high" refer to the overall lighting conditions, considering that AE is performed simultaneously. No additional time is required beyond this photointegration interval, thus significantly reducing the possibility of losing information and generating motion artifacts. After photointegration, the HDR image is immediately ready for processing and rapid decision-making. Furthermore, the present invention can adapt the sensor's operation to any intensity map.This means that the fundamental limitation to achieving a faithful representation of the scene is noise, regardless of lighting conditions. The proposed circuit operates asynchronously once the pixels are reset, so it does not require any external control. All these features are achieved while maintaining an intrinsically linear phototransduction, even when a nonlinear sensor response is implemented overall. The scope of the invention is completed with the method and sensor of the appended claims. Throughout the description and claims, the term "comprises" and its variants are not intended to exclude other technical features, additives, components, or steps. For those skilled in the art, other objects, advantages, and features of the invention will emerge partly from the description and partly from practice.The following examples and drawings are provided by way of illustration and are not intended to restrict the present invention. Additionally, the present invention encompasses all possible combinations of particular and preferred embodiments indicated herein. BRIEF DESCRIPTION OF THE DRAWINGS To complement the description being made and to aid in a better understanding of the features of the invention, a set of drawings is attached as an integral part of said description, in accordance with a preferred example of a practical embodiment thereof, where, for illustrative and non-limiting purposes, the following is represented: Figure 1 shows that the dynamic range of standard CMOS image sensors is approximately 60-70 dB. This range is appropriately exploited by AE techniques, which adapt the sensor's response to specific aspects of the ambient lighting by means of repeated captures.If the sensor's dynamic range is not sufficient to construct an adequate representation of the scene, multiple captures are taken with different exposure times—exposure bracketing—and then combined to produce an HDR image. Figure 2 shows the representation of two linear photointegration processes taking place simultaneously in a pixel. One of these processes is represented by ^. ^^ (^), which evolves according to the lighting from the pixel to the time instant in which the second process, represented by ^ ^ (^), crosses with a reference voltage level indicated with ^ ^^^ , thus establishing the final pixel voltage ^ ^^^ . It should be noted that ^ ^ (^) initially evolves according to the average illumination that impacts the pixels that make up the sensor array (^ ^ ̅ ^). The two possible practical cases are represented: (a) ^ ^^ < ^ ^ ̅ ^ , which means that ^ ^ (^) crosses ^ ^^^ first of all; and (b) ^ ^^ > ^ ^ ̅ ^ , which means that ^ ^^ (^), cross ^ ^^^ First of all, so that from that moment on fits ^ ^ (^) to evolve according to the pixel illumination from then on. Figure 3 shows two examples of the pixel output, described together with equations (3) and (8), represented on a semi-logarithmic scale. The left trace corresponds to an average photocurrent of ^ ^ ̅ ^ = 1 ^^ while the trace on the right corresponds to an average photocurrent of ^ ^ ̅ ^ = 100 ^^. For both curves, ^ has been set ^^^ = ^ ^^^ = 1.8 ^, ^ ^^^ = 0 ^, and ^ ^^^= 0.9 ^. Figure 4 shows the proposed circuit for implementing the basic pixel operation according to the present invention. The coupling of the two photo-integration processes represented by ^ ^^ (^) and ^ ^ (^) is implemented by switches SW1 and SW2 and inverters INV1 and INV2. The voltage ^ ^^ (^) is introduced to maintain a faithful representation of the average illumination, reproduced in ^ ^ (^) over the required interval. Figure 5 shows examples of the pixel output described by equation (11) below, plotted on a semi-logarithmic scale. All three curves were obtained for the same average photocurrent, ^ ^ ̅ ^ = 10 ^^, and the same reset voltages, ^ ^^^^ = ^ ^^^^ = 1.8 ^. The central curve was obtained for ^ ^^^^ = ^ ^^^^ = 0.9 ^; The curve on the left was obtained for ^ ^^^^ = 1.3 ^ and ^ ^^^^= 0.5 ^; the curve on the right was obtained for ^ ^^^^ = 0.8 ^ and ^ ^^^^ = 1.3 ^. Figure 6 shows the transistor-level layout of the pixel implementation proposed in Figure 4. Elements represented with a thicker line can be shared among several pixels to save area and area and power consumption. ^ ^^^^ , ^ ^^^^ , ^ ^^^^ , and ^ ^^^^ are crucial design variables; these must be adjusted appropriately to maximize the range of ^ ^^^ according to equation (11) while ensuring the correct operation of the different circuit blocks, including the reading by the source follower (SF). Figure 7 shows examples of the two possible dynamics for the electrical realization of the pixel proposed in Figure 4: a) temporal evolution and ^ ^ (^) leading to a final pixel voltage described by equation (16); and (b) time evolution of and ^ ^ (^) which leads to a final pixel voltage described by equation (19). PREFERRED EMBODIMENTS OF THE INVENTION The present invention features an imaging method that integrates AE and HDR asynchronously into a single shot while maintaining a linear response of each pixel, and a sensor design that implements the method. Basic Model In the basic model, figure 2 will be considered, which shows two linear photointegration processes taking place simultaneously during image formation: (1) ^ ^^ (^) represents the evolution of the pixel's own voltage, which decreases linearly from the reset voltage ^ ^^^ according to the local photogenerated current ^ ^^ ; and (2) ^ ^(^) represents the evolution of an auxiliary voltage that initially decreases linearly from the reset voltage, according to the average photogenerated current over the entire pixel matrix, indicated as ^ ^ ̅ ^ . How this global photocurrent is detected will be explained in detail below, with reference to Figures 4 or 6. Therefore, ^ ^^ (^) and ^ ^ (^) initially evolve as follows: [Eq.1]: where C is the photointegration capacity, which is the same (in this basic model) for both photointegration processes, without loss of generality. The photointegration processes are coupled through a reference voltage ^ ^^^ , again common to both processes, without loss of generality. Whenever ^ ^ (^) crosses ^ ^^^ , the process represented by ^ ^^ (^) stops and the final pixel value is set to ^ ^^^ . Yes ^ ^^(^) crosses ^ ^^^ before ^ ^ (^), the evolution of ^ ^ (^) changes, decreasing linearly from that crossing point onwards, according to the local photogenerated current. This coupling can be modeled by considering the two possible scenarios represented in Figure 2. The simplest scenario is represented in Fig. 2(a): since ^ ^^ < ^ ^ ̅ ^ , ^ ^ (^) crosses ^ ^^^ First of all, we will denote this instant in time as ^ ^^^^^ , which can be easily derived from Eq. (1) by setting ^ ^ (^) = ^ ^^^ and obtaining the value of ^: [Eq.2]: and replacing this instant of time with the expression ^ ^^ (^) in Eq.1, the following final pixel value is obtained: [Eq.3]: Therefore, according to Eq. (3), pixels with local illumination below the mean illumination will have a final value proportional to that of the local illumination, always in relative terms with respect to the mean illumination. This is a crucial point for performing AE. Differentiating in Eq. (3), we obtain that is, the exchange rate of ^ ^^^ with respect to local illumination is constant. This means that scene information is transferred to the image plane in the same linear manner for all tones below the mean illumination. It should also be noted that if ^ ^^^ It is set at the midpoint of the signal range — we will denote this midpoint as ^ ^^^ —, set ^ ^^ = ^ ^ ̅ ^ in equation (3) leads to ^ ^^ (^ ^^^^^ ) = ^ ^^^; that is, when the local illumination and the average illumination coincide, the final pixel value falls in the center of the pixel's response, regardless of the absolute values ​​of those illuminations. Next, the second scenario illustrated in Figure 2(b) will be analyzed. In this case, since ^ ^^ > ^ ^ ̅ ^ , ^ ^^ (^) crosses ^ ^^^ First of all, we will denote this instant , which can be easily derived from Eq. (1) by establishing the value of ^: [Eq.4]: ^(^ ^^^ − ^ ^^^ ) ^ ^^^^^ = ^ ^^ Substituting this instant of time in the expression ^ ^ (^) in Eq. (1), the following value has been obtained: [Eq.5]: As mentioned above, from ^ ^^^^^ henceforth, ^ ^ (^) decreases linearly according to the local photogenerated current: [Eq.6]: Setting ^ (^) = ^ ^^^in Eq. (6) and obtaining the value of ^, that is, the instant in time at which ^ ^ (^) crosses ^ ^^^ in this second scenario: [Eq.7]: Finally, replacing ^ ^^^^^ in the expression ^ ^^ (^) In Eq. (1), the final value of the pixel has been obtained when ^ ^^ > ^ ^ ̅ ^ : [Ec.8]: Therefore, according to equation (8), pixels with local illumination above the mean illumination will have a final value inversely proportional to the local illumination, again in relative terms with respect to the mean illumination. In this case, ^^ ^^^ ⁄ ^^ ^^ = − ^ ^ ^^^ − ^ ^^^ ^ ^ ^ ̅ ^ ^ ^ ^ ^ ^ , ∀^ ^^ > ^ ^ ̅ ^ , that is, the exchange rate of ^ ^^^With respect to local illumination, it decreases quadratically with local illumination. This means that illumination values ​​above the average illumination will be progressively compressed within the available signal range, to the point where no additional scene information is actually transferred to the image plane, due to noise or the resolution of the readout circuit. It should be noted that for very high pixel illuminations – relative to the average illumination, i.e., ^ ^^ ≫ ^ ^ ̅ ^ — , ^ ^ ̅ ^ / ^ ^^ ≃ 0. In this case, the following equation has been obtained from equation (8): [Eq.9]: If we assume again that ^ ^^^ It is set at the midpoint of the signal range, which can be expressed as ^ ^^^ = (^ ^^^ + ^ ^^^ ) / 2. In our case, we will have to ^ ^^^ = ^ ^^^ . And also, replacing ^ ^^^= (^ ^^^ + ^ ^^^ ) / 2 by Eq. 9, we obtain that: [Eq.10]: In short, it has been concluded that, if ^ ^^^ = ^ ^^^ = (^ ^^^ + ^ ^^^ )⁄ 2 ^ ^ ^^^ = ^ ^^^ , the proposed detection scheme produces a pixel voltage equal to: ^ ^ ^^^ for ^ ^^ ≫ ^ ^ ̅ ^ ; note that equations (3) and (8) converge at ^ ^^^ = Remarkably, this ideal adaptation of the photocurrents in the available signal range, expressed by equations (3) and (8) – linear for low illuminations, compressive for high illuminations – takes place independently of the particular absolute magnitude of the photocurrents, since ^ ^^^ depends on ^ ^^ / ^ ^ ̅ ^ ,or of ^ ^ ̅ ^ / ^ ^^ . Another point to emphasize is that ^ ^^(^) always evolves linearly until the final pixel voltage is established, even though ^ ^^^ present a non-linear relationship with ^ ^^ for^ ^ ^^ > ^ ^^ . Therefore, the proposed coupling between the photointegration processes allows that, although the detection occurring at the pixel is inherently linear, it results in an overall sensor response that is nonlinear. Figure 3 presents two examples of the pixel output (^ ^^^ ) for a sweep of six orders of magnitude in ^ ^^ assuming that ^ ^ ̅ ^ = 1 ^^ (left curve) and ^ ^ ̅ ^ = 100 ^^ (right curve). For both curves, ^ has been set ^^^ = ^ ^^^ = 1.8 ^, ^ ^^^ = 0 ^, and ^ ^^^ = 0.9 ^. Therefore, the average photocurrent is represented by ^ ^^^in both cases. It should be noted that the proposed detection strategy adapts photocurrents that cover a very wide range around the average illumination within the available signal range. Electrical Implementation As mentioned above, the above description refers to the basic model of a pixel according to the present invention. Figure 4 shows a pixel circuit that implements the basic model presented in the previous paragraphs, with some additional attributes as will be explained below. The voltage ^ ^^ (^), evolves linearly from ^ ^^^^ According to the mean illumination, once the reset period ends at ^ = 0 ^ this is the starting point of the global shutter photointegration that leads to image formation. The detection of the mean illumination is achieved by interconnecting the node associated with ^ ^^(^) at each pixel with the corresponding nodes of neighboring pixels to the north, south, east, and west across the entire 2D array. This results in a grid on which the pixels are interconnected, and at the same time the average of the photocurrents generated in the photodiodes with area ^ / ^ is calculated through a charge redistribution in the capacitors with a capacity equal to ^ / ^. Note that the op-amp, connected in a buffer configuration, copies ^ ^^ (^) in ^ ^ (^) until ^ ^^ (^) crosses the threshold voltage of INV1, indicated as ^ ^^^^ At this instant in time, the switch SW1, controlled by the output of the inverter INV1, disconnects the output of the operational amplifier from ^ ^ (^), which continues to evolve from that moment onwards according to the current generated by the photodiode with area ^ / ^. Similarly, when ^ ^ (^), crosses the threshold voltage of INV2, indicated as ^^^^^ , the SW2 switch disconnects the photodiode with area ^ from the sensing capacitor, establishing the final pixel voltage. The fact that two different reset voltages have been considered —^ ^^^^ and ^ ^^^^ — and two different reference voltages —^ ^^^^ and ^ ^^^^ — make the derivation of ^ ^^^ is slightly more advanced than the basic mode described above. In the section “Analytical model of pixel realization” it is shown that ^ ^^^ For the circuit proposed in Figure 4, it is expressed as follows: [Eq.11]: Note that the expressions in Eq. (11) converge in equations (3) and (8), respectively, if ^ ^^^^ = ^ ^^^^ = ^ ^^^ and ^ ^^^^ = ^ ^^^^ = ^ ^^^ . The availability of ^ ^^^^ , ^ ^^^^ , ^ ^^^^ and ^ ^^^^as design parameters provides the pixel realization with flexibility to shape the sensor response. Figure 5 shows three pixel output curves described by equation (11) for ^ ^ ̅ ^ = 10 ^^, ^ ^^^^ = ^ ^^^^ = 1.8 ^, and different values ​​of ^ ^^^^ and ^ ^^^^ . Interestingly, changes in these reference voltages imply a significant variation in how photocurrents near the average illumination map onto the available signal range. As a final comment, the parameters ^ and ^ in Figure 4 are assumed to be greater than unity and reflect the fact that, because the photointegration processes associated with ^ ^ (^) and ^ ^^(^) are auxiliary in the generation of the final pixel value, the areas of the corresponding photodiodes and sensing capacitors can be proportionally reduced in order to increase the fill factor and reduce the pixel size (pixel pitch), which are two fundamental parameters of image sensors. In natural vision systems, the retina is a specialized tissue at the front of the system that detects light and performs the initial processing tasks before sending an electrical message through the optic nerve to the brain for further processing and completion of visual perception. Automatic light adaptation is one of those initial tasks. Although the proposed circuit does not implement any of the extremely complex signaling pathways identified in the retina, there are general characteristics that do coincide.First, in the retina, as in our approach, adaptation is related to the average illumination level and relies on appropriate tuning of the temporal response of photoreceptors. The range of this response is narrow in both cases and cannot encompass the wide variety of illuminations observed in real-life scenarios. In the retina, the average illumination is computed by horizontal cells (HCs), which average cone activity through strong electrical coupling. HCs, which constitute the closest network of cells interacting with cones, also provide receptors with highly "personalized" feedback to modulate their response. In the proposed circuit, the average illumination is obtained by averaging the local photoresponse across the pixel array using charge redistribution, i.e., strong electrical coupling.Furthermore, similar to HCs, the same circuit elements that sense the average illumination provide feedback to each pixel's photointegration process to appropriately modulate its output according to the ambient lighting conditions. Pixel Layout Figure 6 shows a transistor-level layout of the pixel implementation proposed in Figure 4. It is implemented in a standard 0.18μm 3.3V CMOS technology. The thicker trace elements can be shared among multiple pixels to save area and power consumption, as will be described later. For now, each pixel is assumed to include the circuitry depicted in Figure 6. According to equation (11), ^. ^^^ It is located in a range that goes from a maximum value of ^ ^^^^ for ^ ^^ ≪ ^ ^ ̅ ^ up to a minimum value of ^^ ^^^^ + ^ ^^^^ − ^ ^^^^ ^ for ^ ^^ ≫ ^ ^ ̅^ . To extend the signal range of the pixel as much as possible at the upper limit, it is clear that ^ ^^^^ should be set to ^^^ = 3.3 ^. As far as the minimum of ^ is concerned ^^^ , must be set to a value such that the output through the source follower (SF) can be realized appropriately. Therefore, the inverters for threshold voltages have been designed to be ^ ^^^^ = ^ ^^^^ = 2.15 ^, thus allowing that ^ ^^^^ is set to ^^^ = 3.3 ^ as well, so that the minimum value of ^ ^^^ is nominally equal to 1 V, high enough for the source follower to provide a ^ ^^^^^ suitable. To set this way ^ ^^^^ , ^ ^^^^ , ^ ^^^^ and ^ ^^^^ , the transition between the two response curves in equation (11) takes place at ^ ^^ = ^ ^ ̅ ^ ; at this point, ^ ^^^ = ^ ^^^= 2.15 ^. The photodiodes were simulated using a Verilog-AMS model. In this model, the incident optical power is set as a parameter to electrically simulate the sensor response to HDR scenes under different ambient illuminations, as will be described in the “Analytical model of the pixel realization” section. It should be noted that ^ = ^ = 2 in the pixel layout shown in Fig. 6. This downscaling of the auxiliary photodiodes and the corresponding sensing capacitors could be more aggressive. However, a decision has been made to be conservative in this regard to ensure the targeted functionality. As a final comment, the NMOS-based capacitor realization aims to save area at the expense of introducing higher nonlinearities with respect to other solutions, e.g., MIM capacitors. Analytical model of the pixel realization For the pixel realization shown in Fig.4, t = 0 sets the end of the reset interval and the beginning of the image formation process. A global shutter is performed for proper detection of ambient illumination. After reset, ^. ^^ (^) is set to ^ ^^^^ and ^ ^^ (^) to ^ ^^^^ . These voltages must be greater than ^ ^^^^ and ^ ^^^^ , respectively. Thus, the output of inverters INV1 and INV2 will be set to logic “0” at t = 0; switches SW1 and SW2 must be designed to be ON for this control signal, thus establishing an electrical connection between the corresponding circuit nodes. For SW1, this means that ^ ^^ (^) is initially copied to ^ ^ (^); for SW2, it means that the sensing capacitance ^ is initially connected to the area photodiode ^. Since, as mentioned in the electrical realization section, ^ ^^(^) represents the average illumination incident on the 2D pixel array, ^ ^^ (^) and ^ ^ (^) initially evolve as follows: [Eq.12]: The expressions in equation (12) will change when ^ ^^ (^) crossing ^ ^^^^ or when ^ ^ (^) crossing ^ ^^^^ , in this way SW1 or SW2 will become OFF, respectively. It has been assumed, as a first case, that ^ ^ (^) crosses ^ ^^^^ Firstly at an instant in time which, by analogy with the derivations in the basic model, is indicated as . This time instant can be easily obtained from equation (12) by setting ^ ^ (^) = ^ ^^^^ and obtaining ^: [Eq.13]: In ^ = ^ ^^^^^ , the output of INV2 changes to logic “1” and SW2 is OFF, setting the final value of the pixel ^ ^^^in the sensing capacity C. Before replacing equation (13) by the expression ^ ^^ (^) in equation (12), the condition that leads to ^ must be analyzed ^ (^) to cross ^ ^^^^ before ^ ^^ (^) crossing ^ ^^^^ . According to equation (12), ^ ^^ (^) would cross ^ ^^^^ in an instant of time ^ ^^^^^ —using again the basic model notation— expressed as follows: [Eq.14]: Therefore, the condition for this first case to occur is that < ^ ^^^^^ , which after applying some algebra concepts produces: [Eq.15]: Finally, substituting equation (13) into the expression for ^ ^^ (^) in equation (12), we have obtained: [Eq.16]: An example of this first dynamic is represented in Figure 7(a). Note that not only the relative positions of ^ ^^^^ and ^ ^^^^are relevant in these dynamics, but so are those of ^ ^^^^ and ^ ^^^^ , in addition to the difference between the local photocurrent ^ ^^ and the average illumination encoded by ^ ^ ̅ ^ The second possible case occurs when . In this case, it happens that^ ^^ (^) crosses ^ ^ first of all. this way, in ^ = ^ ^^^^^ , the output of INV1 changes to logic “1” and the switch SW1 is OFF. Consequently, from ^ = ^ ^^^^^ henceforth, ^ ^ (^) evolves according to ^ ^^ as follows, [Ec.17]: When the time comes, ^ ^ (^), following the dynamics described by equation (17), will cross ^ ^^^^ , setting the final value of the pixel in this second case. Thus, if we do ^ ^ (^) = ^ ^^^^ in equation (17) and we calculate ^ we arrive at: [Eq.18]: Finally, replacing ^ = ^ ^^^^^in the expression ^ ^^ (^) described by equation (12), we obtain: [Eq.19]: An example of this second dynamic is shown in Fig. 7(b). Note that, in both possible dynamics, ^ ^^ (^) evolves linearly until the final pixel value is established. Finally, it should also be noted that equations (16) and (19) converge

Claims

CLAIMS 1. A pixel operation method comprising two linear photointegration processes taking place simultaneously during the formation of an image, wherein: the first photointegration process represents the evolution of the voltage of the pixel itself (^ ^^ (^)), which decreases linearly from the reset voltage (^ ^^^ ) according to the photogenerated current (^ ^^ ); and the second linear photointegration process represents the evolution of an auxiliary voltage (^ ^ (^)) which initially decreases linearly from the reset voltage (^ ^^^ ) according to the average photogenerated current across the entire pixel array characterized in that the first and second photointegration processes are coupled through a reference voltage (^ ^^^) common to both the first and second photointegration processes, such that whenever the auxiliary voltage (^ ^ (^)) crosses the reference voltage (^ ^^^ ), the first linear photointegration process stops, and a final pixel value is set in which: (a) if the auxiliary voltage (^ ^ (^)) crosses the reference voltage (^ ^^^ ) firstly, consequently the local photogenerated current is lower than the average photogenerated current over the entire pixel array (^ ^^ < ^ ^ ̅ ^ ), so that said pixel receives a local illumination below the average illumination, and the rate of change of the final value of the pixel with respect to the local illumination is constant; (b) if the pixel voltage (^ ^^ (^)) crosses the reference voltage (^ ^^^) firstly, consequently the local photogenerated current is greater than the average photogenerated current over the entire pixel array (^ ^^ > ^ ^ ̅ ^ ) establishing that ^ ^ (^) evolve according to the pixel's illumination from an instant forward; said pixel presents a local illumination above the average illumination, and the rate of change of the final value of the pixel (^ ^^^ ) with respect to local illumination decreases quadratically with local illumination.

2. The pixel operation method according to claim 1, wherein when the rate of change of the final pixel value (^ ^^^) with respect to local illumination is constant, the scene information is transferred to the image plane in the same linear manner for all tones below the average illumination.

3. The pixel operation method according to claim 1 or 2, wherein by decreasing the rate of change of the final pixel value (^ ^^^) with respect to the local illumination quadratically with the local illumination, illumination values above the mean illumination will be progressively compressed into the available signal range until no scene information is transferred to the image plane due to noise or the resolution of the readout circuit.

4. A pixel array circuit arranged to operate asynchronously once the pixels are reset, without requiring any external control according to the pixel operation method according to any one of claims 1 to 3, comprising a first inverter (INV1) and a second inverter (INV2); an operational amplifier in a buffer configuration; a first photodiode with an area ^, a second photodiode with an area ^ / ^ and a third photodiode with an area ^ / ^, where ^ and ^ are parameters greater than unity, characterized in that the mean illumination voltage (^ ^^(^)) evolves linearly from a second reset voltage (^ ^^^^ ) according to the average illumination once the reset period is over, which is the starting point of the global shutter photointegration leading to the image formation; where the detection of the average illumination is achieved by interconnecting a node associated with the average illumination voltage (^ ^^ (^)) at each pixel with corresponding nodes of neighboring pixels located to the north, south, east, and west along a 2D array, arranged in a grid in which all pixels are interconnected; wherein the average of the photocurrents generated in the photodiodes with an area ^ / ^ is calculated simultaneously through the redistribution of the charge in a first capacitor (^ / ^) and the operational amplifier in a buffer configuration copies the average illumination voltage (^ ^^ (^)) in the auxiliary voltage (^ ^(^)) until the pixel voltage (^ ^^ (^)) crosses a first threshold voltage (^ ^^^^ ) of the first investor (INV1), and at this moment, a first key (SW1) that is controlled by the output of the first inverter (INV1) disconnects the output of the operational amplifier connected as a buffer from the auxiliary voltage (^ ^ (^)) which continues to evolve from that instant onwards, according to the local photocurrent generated by the photodiode with an area ^ / ^; and where, when the auxiliary voltage (^ ^ (^)) crosses the second threshold voltage (^ ^^^^ ) of the second inverter (INV2), the second switch (SW2) disconnects the first photodiode with an area ^ of its sensing capacitor (^), setting the voltage of the final value of the pixel 5. A system for simultaneous automatic exposure (AE) and high dynamic range (HDR) imaging in a single shot, characterized in that it simultaneously performs AE and HDR during image capture; and wherein for each photointegration period set to produce the image, a circuit of a pixel array according to claim 4 automatically adjusts its response to an average illumination at the time and adapts a radiation map of the scene within a range of available pixel signal.

Images