On-board real-time tracking of pedestrians on a UAV

Abstract

Recent technical advances in Unmanned Aerial Vehicles (UAV) made a realm of applications possible. In this paper we focus on the application of following a walking pedestrian in real-time, using optimised pedestrian detection and object tracking. For this we use an on-board embedded system, offering an optimal ratio of computational power and weight. We extend the commonly used ground plane estimation technique, used to reduce the search space, based on the sensor data off the UAV. The integration of the ground plane constraint obtains a significant speed-up over the already optimised Aggregate Channel Feature (ACF) detector. To compensate for the frames without detections, we use a particle tracker based on color information. We successfully validated our system on a flying UAV.

Remote Estimation of Target Height from Unmanned Aerial Vehicle (UAV) Images

“…Therefore, considering just κ 1 in (12) is deemed enough for this analysis. r corr calculated with (12). needs to be denormalized to obtain the undistorted µ i−corr and ν i−corr image coordinates for the image under study.…”

“…Moreover,r i is a distorted value that needs to be compensated to obtain the distance r i−corr of the ideal undistorted image. The procedure to obtain such undistorted distance was already discussed in the previous paragraph (see (12)). Similarly, Image Point J is the intersection of the image plane with the ray line that connects the bottom of the pole (lowest point) with the camera center.…”

“…where r i−corr can be calculated considering (12). in the previous paragraph starting from the observed r i in the image plane (see Figure 6).…”

“…The total number of pixels spanning the entire pole after distortion correction (r i−corr + r j−corr ) as well as the calculated elevation of the UAV (defined using the formula (17)) are also reported in Table 2. The Distortion Coefficient to be used for the correction was retrieved through camera calibration techniques [15] developed with OpenCV via Python programming and is equal to 0.0275 (this is the k1 coefficient as expressed in the model (12)). Please see Section 2.2 for further clarification.…”

“…The widely used approach of using vanishing lines with a reference object present in the scene for height estimation has relevant setbacks: defining vanishing lines and reference objects may not be always possible in an image. On the other hand, the pinhole model does not require vanishing lines [11] and therefore is deemed very suitable for UAV operations [12]. Photogrammetric techniques require having either a double camera pointing at the same target or acquiring at least two images from different orientations of a (static) feature.…”