Water Hazard Detection And Tracking For UGV In Off-road Environment Based On SVM And SURF

Unmanned ground vehicle (UGV) belongs to a typical type of intelligent mobile robots who are capable to recognize travelable region based on environmental information collected by robots itself and design an optimal path without a driver’s interference. Unmanned technology utilizes variety of sensors, data processing device and action execution mechanism by combining state of art technology from multi-discipline. Environment perception system, takes charge of collecting and processing environmental information, regarded as the fore-end module of the entire unmanned system and its performance directly affects whether driving task can be achieved independently.Obstacle detection known as the key part of the environment perception system, while detecting obstacle in unstructured environments faces enormous challenge for mess background and varied types of obstacles. In this paper, we focuse on the new field about detection and tracking of wild water body obstacle, what’s more we conduct research to explore the salient features extracted from visual information of water region collected by passive sensors. Generalization ability of the algorithm could be improved by adding support vector machine(SVM). Finally,we proposed a water obstacle detection and tracking method based on SVM and SURF.The specific contents are as follows.(1) As the performance of traditional water detection approach by using single brightness feature seems unsatisfied enough, we presents a brand new method combine S/V color feature extracted by HSV converted from RGB and ASM (Angular Second Moment), ENT (Entropy), CON (Contrast) and COR (Correlation) texture feature calculated by GLCM (Gray-level co-occurrence matrix) to build a5-D descriptors to describe water feature.(2) For nearly each individual feature from the5-D descriptors, there is a relatively obvious boundary exists between feature stands for water region and terrain surface around respectively. So we consider training the feature statistics to adapt more complex environment. SVM (support vector machine) developed from basic statistical learning theory while performs excellent in solving nonlinear problems and limited sample problem. Color and texture information extracted from image pixels are utilized to form a feature matrix to train the SVM classify process. While, with the assistance of RBF kernel function, low-dimensional sample space is projected to high-dimensional space. Based on large amount of experiments, RBF kernel function parameters are optimized by using grid method. The optimized RBF kernel function parameters are proved their satisfied effect in the static water detection experiments.(3) For water hazard tracking, we proposed an approach based on SURF detection algorithm to search salient features from water region. In this paper, SURF establishes the relationship of feature points within the ROI (region of interest) between adjacent frames and match them to calculate the feature points’ changing on position and scale, which guide us to update the position and scale of ROI. SURF is to match local features with scale and rotational invariance, thus have a good adaptability dealing with zooming, changing of viewing Angle, noise or occlusion. What’s more, SURF algorithm has been greatly improved over the SIFT algorithm in the algorithm execution time confirmed by experiment.(4) Finally, according to the foregoing findings, the SVM water detection and SURF feature tracking algorithm are organically combined to form a complete system through the matching technique. While terrain surface condition rugged makes the CCD shaking significantly, which impact the water region tracking effect. Thus, in this paper we proposed solutions to two typical type of water body tracking failure modes. Finally, water hazard tracking experiment are carried out under different lighting conditions. Experimental results indicate that the algorithm proposed in our paper could achieve effective detection and tracking of water hazard in mid-range distance to the front of the vehicle.