Granted Patent B2

US 12,732,717 · App. 18/728,428 · Granted Sep 8, 2026

Inventors: Ward Van Der Tempel (Keerbergen, BE); Christian Mourad (Genval, BE)

Assignee: VoxelSensors SRL

H04N25/77H04N25/40

Abstract

Claims (34)

1 . An image sensor for efficient optical sensing, comprising:

a plurality of pixel sensors, each pixel sensor comprising a photo detector, and being adapted to output a logic signal, and

processor, wherein the processor is at least one neural network, comprising at least one layer of neurons,

each photo detector is stacked on at least one neuron corresponding thereto of the layer, wherein each photo detector is connected in parallel to the at least one neuron corresponding thereto of the layer,

characterized in that the photo detector is a single-photon detector and that each neuron is adapted to receive a train of pulses, wherein each neuron is connected to at least one neighboring neuron, wherein each neuron is capable of producing an output based on input signals from the at least one neighboring neuron and the output of the detector, wherein the neural network is capable of:

identifying detections connected or semi-connected or exhibiting a specific pattern in space over consecutive time steps, or

identifying at least two neighboring detections over one time step.

2 . An image sensor of claim 1 , wherein each neuron is capable of producing an output based on current inputs to the neuron as well as current state of the neuron.

3 . An image sensor of claim 1 , wherein the neural network is a spiking neural network.

4 . An image sensor of claim 1 , wherein a distance between each detector and the processor is at most 100 micrometers.

5 . The image sensor of claim 4 , wherein the distance between each detector and the processor is at most 50 micrometers.

6 . An optical sensing system comprising an image sensor of claim 1 and optics able to produce an image of a scene on the image sensor.

7 . An optical sensing system of claim 6 , wherein the system further comprises at least one light source, wherein the at least one light source is adapted to illuminate a dot on a scene, wherein at least one light source is adapted to scan the dot on the scene.

8 . The optical sensing system of claim 7 , wherein the at least one light source is adapted to scan the dot on the scene continuously.

9 . An optical sensing system of claim 6 , wherein the network is trained to extract features of a set of input signals to the network.

10 . A method for extracting features of a scene using the optical sensing system of claim 6 , comprising the steps of:

adapting the neural network to receive a train of pulses,

adjusting an activation level of a neuron after receiving each pulse by the neuron,

adapting the neuron to produce an output signal once the activation level exceeds a threshold value,

propagating the output signal to at least one neuron in a subsequent layer of the network, and

extracting features of output signals of the network.

11 . A method of claim 10 , wherein the method further comprises the step of determining the depth profile of a field of view, the determining step comprises the steps of:

projecting by the light source at least one light pattern onto the field of view, the projection occurring in a time window of less than 10 μsec;

imaging the projected pattern by the image sensor and optics, synchronized with the projection of the pattern during at least one observation window within the time window, the pixel sensors being in a false status when no light is detected by the corresponding photodetector, and in a true status when light is detected by the corresponding photodetector thereby obtaining a first binary matrix of pixels representing the field of view;

separating the projected pattern from an ambient light noise on the binary matrix of pixels by considering only pixels in the true status having at least one neighboring pixel also in the true status on the matrix of pixel obtained in one observation or a combination of the at least two observations within the time window thereby obtaining a second binary matrix of pixels representing the projected pattern;

based on triangulation between the light source position, the image sensor position and the second binary matrix of pixels, calculating the depth profile corresponding to the projected pattern;

scanning the projected pattern by repeating the projecting, imaging, separating, and calculating steps on the entire field of view for determining the depth profile on the entire field of view;

wherein each isolated element of the pattern is extending in the binary representation on at least two contiguous pixels.

12 . The method of claim 11 , wherein the project occurs in a time window of less than 1 μsec.

13 . A method of claim 10 , wherein the method further comprises the steps of:

coding intensity and change of intensity data of impinging photons in the time-domain of each detector, and

decoding the intensity and change of intensity data by the neural network.

14 . The image sensor of claim 1 , wherein each photodetector is directly connected to the at least one neuron corresponding thereto.

15 . The image sensor of claim 1 , wherein the single-photon detector is a single-photon avalanche diode.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (19)