Electronic image stabilization to improve video analytics accuracy
An apparatus includes an image sensor and a processor. The image sensor may be configured to generate a sequence of video frames based on a targeted view of an environment. The processor may be configured to (A) perform an image stabilization on the sequence of video frames and (B) generate video analytics for the sequence of video frames, wherein the image stabilization comprises maintaining a horizon of the targeted view and is performed prior to generating the video analytics.
1. An apparatus comprising:
an image sensor configured to generate a sequence of video frames based on a targeted view of an environment; and
a processor configured to (i) collect data samples from a gravity sensor (G-sensor) and a gyroscopic sensor, (ii) detect X/Y/Z axes transitions using the data samples from said gyroscopic sensor, (iii) perform, in real-time, electronic image stabilization comprising (a) matching axes transition samples with captured image data of the sequence of video frames and (b) generating stabilized captured image data by compensating the captured image data for X/Y axes transitions using an X/Y offset and compensating the captured image data for Z axis transitions using a Z offset, (iv) compute a horizon level based on the data samples from said G-sensor and the data samples from said gyroscopic sensor, (v) perform, in real-time, a rotational offset compensation on the stabilized captured image data of the sequence of video frames to maintain alignment of a respective horizon in each frame of the sequence of video frames with the computed horizon level, and (vi) generate video analytics for the sequence of video frames using the compensated and stabilized captured image data, wherein the electronic image stabilization and rotational offset compensation provides improved accuracy of the video analytics generated for the sequence of video frames.
2. The apparatus according to claim 1 , further comprising one or more sensors configured to capture movement information used to determine the electronic image stabilization applied to the captured image data prior to the rotational offset compensation.
3. The apparatus according to claim 2 , wherein said one or more sensors comprise at least one of an inertial measurement unit and an accelerometer.
4. The apparatus according to claim 1 , wherein said video analytics comprise face detection.
5. The apparatus according to claim 1 , wherein said video analytics comprise face tracking.
6. The apparatus according to claim 1 , wherein said apparatus is part of an unmanned aerial vehicle.
7. The apparatus according to claim 1 , wherein (i) said processor comprises a digital signal processing hardware module and (ii) said electronic image stabilization is performed on said captured image data by said digital signal processing hardware module prior to the rotational offset compensation.
8. The apparatus according to claim 7 , wherein said digital signal processing hardware module is further configured to perform said rotational offset compensation on said stabilized captured image data to maintain said horizon level of said targeted view.
9. The apparatus according to claim 1 , wherein said apparatus is implemented as a system on chip (SoC).
10. A method of performing video analytics comprising:
generating a sequence of video frames based on a targeted view of an environment using a video camera;
collecting, using a processor, data samples from a gravity sensor (G-sensor) and a gyroscopic sensor;
detecting X/Y/Z axes transitions, using said processor, based on the data samples from said gyroscopic sensor;
performing, in real-time using said processor, electronic image stabilization comprising (i) matching axes transition samples with captured image data of the sequence of video frames and (ii) generating stabilized captured image data by compensating the captured image data for X/Y axes transitions using an X/Y offset and compensating the captured image data for Z axis transitions using a Z offset;
computing, using said processor, a horizon level based on the data samples collected from said G-sensor and the data samples collected from said gyroscopic sensor;
performing, in real-time using said processor, a rotational offset compensation on the stabilized captured image data of the sequence of video frames to maintain alignment of a respective horizon in each frame of the sequence of video frames with the computed horizon level; and
generating video analytics for the sequence of video frames using the compensated and stabilized captured image data, wherein the electronic image stabilization and rotational offset compensation provides improved accuracy of the video analytics generated for the sequence of video frames.
11. The method according to claim 10 , wherein said video analytics comprise face detection.
12. The method according to claim 10 , wherein said video analytics comprise face tracking.
13. The method according to claim 10 , wherein said video camera is part of an unmanned aerial vehicle.
14. The method according to claim 10 , further comprising performing vibration correction on the sequence of video frames.
15. The method according to claim 10 , further comprising using one or more sensors to capture movement information used to determine the electronic image stabilization applied to the captured image data prior to the rotational offset compensation.
16. The method according to claim 15 , wherein said one or more sensors comprise at least one of an inertial measurement unit and an accelerometer.