Enhanced face-detection and face-tracking for resource-limited embedded vision systems
Embodiments described herein provide various examples of a real-time face-detection, face-tracking, and face-pose-selection subsystem within an embedded vision system. In one aspect, a process for identifying near-duplicate-face images using this subsystem is disclosed. This process includes the steps of: receiving a determined best-pose-face image associated with a tracked face when the tracked face is determined to be lost; extracting an image feature from the best-pose-face image; computing a set of similarity values between the extracted image feature and each of a set of stored image features in a feature buffer, wherein the set of stored image features are extracted from a set of previously transmitted best-pose-face images; determining if any of the computed similarity values is above a predetermined threshold; and if no computed similarity value is above the predetermined threshold, transmitting the best-pose-face image to a server and storing the extracted image feature into the feature buffer.
1. A computer-implemented method for identifying near-duplicate face images and selectively transmitting best-pose-face images to a server, the method comprising:
receiving a determined best-pose-face image associated with a tracked face when the tracked face is determined to be lost;
extracting an image feature from the best-pose-face image;
computing a set of similarity values between the extracted image feature and each of a set of stored image features in a feature buffer, wherein the set of stored image features are extracted from a set of previously transmitted best-pose-face images;
determining if any of the computed similarity values is above a predetermined threshold; and
if no computed similarity value is above the predetermined threshold,
transmitting the best-pose-face image to a server; and
storing the extracted image feature into the feature buffer.
2. The computer-implemented method of claim 1 , wherein the tracked face is determined to be lost when the tracked face is blocked by an object for a predetermined number of video frames.
3. The computer-implemented method of claim 1 , wherein the image feature is one of or a combination of: a histogram of oriented gradients (HoG) feature, a Harr-like feature, a scale-invariant-feature transform (SIFT)-feature, and a deep-learning-based face feature.
4. The computer-implemented method of claim 1 , wherein the set of stored image features is the same type of feature as the extracted image feature.
5. The computer-implemented method of claim 1 , wherein computing a similarity value between the extracted image feature and a stored image feature in the set of stored image features includes computing a cosine similarity between the extracted image feature and the stored image feature.
6. The computer-implemented method of claim 1 , wherein if the image feature is a deep learning (DL)-based face feature, comparing the extracted image feature with the set of stored image features allows for determining if the tracked face is a duplicated face of a stored face of the same person but having a different pose from the stored face.
7. The computer-implemented method of claim 1 , wherein if any of the computed similarity values is above the predetermined threshold, the method further comprises preventing transmitting the best-pose-face image to the server.
8. An embedded system capable of identifying near-duplicate-face images and selectively transmitting best-pose-face images to a server, the embedded system comprising:
a processor; and
a memory coupled to the processor,
wherein the memory storing instructions that, when executed by the one or more processors, cause the system to:
receive a determined best-pose-face image associated with a tracked face when the tracked face is determined to be lost;
extract an image feature from the best-pose-face image;
compute a set of similarity values between the extracted image feature and each of a set of stored image features in a feature buffer, wherein the set of stored image features are extracted from a set of previously transmitted best-pose-face images;
determine if any of the computed similarity values is above a predetermined threshold; and
if no computed similarity value is above the predetermined threshold,
transmit the best-pose-face image to a server; and
store the extracted image feature into the feature buffer.
9. The embedded system of claim 8 , wherein the tracked face is determined to be lost when the tracked face is blocked by an object for a predetermined number of video frames.
10. The embedded system of claim 8 , wherein the image feature is one or a combination of: a histogram of oriented gradients (HoG) feature, a Harr-like feature, a scale-invariant-feature transform (SIFT)-feature, and a deep-learning-based face feature.
11. The embedded system of claim 8 , wherein computing a similarity value between the extracted image feature and a stored image feature in the set of stored image features includes computing a cosine similarity between the extracted image feature and the stored image feature.
12. The embedded system of claim 8 , where if the image feature is a DL-based face feature, comparing the extracted image feature with the set of stored image features allows for determining if the tracked face is a duplicated face of a stored face of the same person but having a different pose from the stored face.
13. The embedded system of claim 8 , wherein if any of the computed similarity values is above the predetermined threshold, the system prevents transmitting the best-pose-face image to the serve.
14. The embedded system of claim 8 , wherein the embedded system is one of: a surveillance camera system, a machine vision system, a drone system, a robotic system, a self-driving vehicle, and a mobile device.