Unmanned Aerial Vehicles have been used widely in the commercial and surveillance use in the recent year. Vehicle tracking from aerial video is one of commonly used application. In this paper, a self-learning mechanism has been proposed for the vehicle tracking in real time. The main contribution of this paper is that the proposed system can automatic detect and track multiple vehicles with a self-learning process leading to enhance the tracking and detection accuracy. Two detection methods have been used for the detection. The Features from Accelerated Segment Test (FAST) with Histograms of Oriented Gradient (HoG) method and the HSV colour feature with Grey Level Co-occurrence Matrix (GLCM) method have been proposed for the vehicle detection. A Forward and Backward Tracking (FBT) mechanism has been employed for the vehicle tracking. The main purpose of this research is to increase the vehicle detection accuracy by using the tracking results and the learning process, which can monitor the detection and tracking performance by using their outputs. Videos captured from UAVs have been used to evaluate the performance of the proposed method. According to the results, the proposed learning system can increase the detection performance.<\/p>\n\n\n\n
Unmanned Aerial Vehicles\n(UAVs) have become a key research area in recent years in military and civilian\napplications, which has the advantages of small, lightweight, fast and easy\ndeployment, as well as it can achieve \u201czero\u201d casualties so it can be deploy at\nthe extreme missions. Vehicle detection from UAVs has drawn a great attention\nin the researches such as automatic traffic monitoring, aerial surveillance and\nother security related applications. There are various challenges that UAVs\ncould face, one of the main challenges of the detection and tracking is the\ntarget objects might change their shapes in the aerial images or sudden\ndisappear and reappear during the tracking process. Thus, the detection and\ntracking process needs to handle various problems. First of all, the tracking\nand detection system have to be scale-invariant to the target which avoid the\nerrors caused by the UAVs changing their altitude during tracking. Secondly,\nthe rotationally invariant features should be considered as the UAV\u2019s flight\ndirections can change rapidly and unpredictable, which change the directions of\nthe target\u2019s movement. Furthermore, the illumination to the targets may vary\ndepending on the flight directions of the UAVs and shooting angles to the\ntargets, also, the blur problems could occurred by the UAVs\u2019 shaking.\nTherefore, the transformation invariant is needed. Furthermore, the background\nconfusions and targets occlusions may exist. Finally, the most important issue\nis the detection and tracking process have to be real-time. In this paper, a\nvehicle tracking and detection method with self-learning has been proposed as\nshown in Figure 1. In the input video, vehicles are detected automatically\nusing the features extracted from Histogram of Oriented Gradients (HoG) [1] and\nFeatures from Accelerated Segment Test (FAST) [2] with Support Vector Machine\n(SVM) classifier [3]. It is assumed that the vehicle has higher density of\ncorners than other objects in the environment so finding the distribution of\ncorners should be the very first thing to narrow the area for further HoG\nprocessing. FAST corner detection method can quickly and accurately detect\nrelevant corner points. Another detection method by using the Grey Level\nCo-occurrence Matrix (GLCM) with HSV colour feature has been used in order to\nprove that the proposed self-learning tracking method can increase the\ndetection accuracy.<\/p>\n\n\n\n
The proposed self-learning tracking\nsystem was inspired by the method of Tracking Learning Detection (TLD) in [4],\nwhich the TLD can track a single target. The proposed approach in this paper\nupgraded it to track multiple targets in real-time. It is assumed that both\ndetection and tracking process could make errors so it is necessary to let them\nmonitor each other. The TLD algorithm can monitor the tracking results by the\ndetection results. In this paper, a Forward and Backward Tracking (FBT)\nmechanism has been proposed, which can self-check whether there is any errors\nin the tracking process by using the previous tracking results. Also, the FBT\ncould monitor the detection results by comparing the similarity with the\ntracking results using the Scalar Invariant Feature Transform (SIFT) feature.\nThe inspectors (positive and negative) have been developed for the error\nestimations. Furthermore, the FBT will also update the classification based on\nthe tracking results for future detection use. Two measures have used for the\nFBT monitoring process. Firstly, it is assumed that when tracking a same target\nin a sequence frames, the features of the target should be very similar. Thus,\nFBT uses SIFT matching process in the tracking results along the tracking\nsequence. If the matching score between the current result and previous result\nis above the threshold, the FBT is tracking the same target. Conversely, if the\nmatching score is below the threshold, the FBT will be considered another\ntarget has been tracked. The FBT has a tracked vehicle database (TVD) which\nstores the SIFT information about already tracked vehicles. The second measure\nis that when the FBT could not match with any targets in the TVD, the FBT will\nconsidered this result as a false positive. Once the FBT gives the tracking\nresult, it will compare with the detection results in the following frame,\nwhich can decide whether the detection result is correct or not. All decisions\nwill be saved to update the classification model for further detections. The\nSIFT matching method has been used because it has a considerable high matching\nperformance with acceptable processing resources requirement. In the TVD, each\nvehicle has its own SIFT points\u2019 descriptors which will be used in the matching\nprocess.<\/p>\n","protected":false},"excerpt":{"rendered":"