Detection of fast incoming objects with a moving camera

 

Abstract

 

The use of a monocular camera for early collision detection to elude fast incoming objects in cluttered scenes is a desirable but very challenging functionality for mobile robots, such as small drones. We present a novel moving object detection and avoidance algorithm for an uncalibrated camera that uses only the optical flow to predict collisions. After camera-motion compensation, we remove the noise in the optical flow by adaptively learning the background motion to reduce the disturbances caused by lens distortion, close terrain and dynamic textures when detecting moving objects. Next, we compute the proximity of an incoming object based on its time to contact, i.e. the time until the object crosses the infinite extension of the image plane. Finally, we combine in a Bayesian framework the compensated motion and the time to contact in order to identify on the image plane an object-free region the robot can move towards to avoid the collision. We demonstrate and evaluate the effectiveness of the proposed algorithm using footage of flying robots that observe fast incoming objects such as birds, balls and other drones.

Reference

 

F. Poiesi A. Cavallaro, "Detection of fast incoming objects with a moving camera," British Machine Vision Conference (BMVC), York, UK, Sep 2016 pdf

Code

 

Matlab code for the method presented in this paper Code

Results

 

Optical flow results

Red arrows: motion vectors

Optical flow results. Note the cases when the optical flow is highly noisy and the object motion is barely visible.

 

 

True Positive, False Positive, False Negative detections

True positives: green bounding box

False positives: yellow bounding box

False negative: red bounding box

True, false and missed detections of incoming objects. False positives occur when static objects are close to the (moving) camera or when texture-less regions produce erroneous optical flow. False negatives are often due to barely visible motion of the incoming object.

 

 

Safe point localization

Green cross: safe point

The safe point is the estimated location in the image a flying robot could exploit to avoid and incoming object. The blue areas are the minima of the posterior distribution and represent regions that could contain incoming objects.

Acknowledgments

 

This work was supported in part by the Artemis JU and in part the UK Technology Strategy Board through COPCAMS Project under Grant 332913.

Contact