Moving Vehicle Detection Based On Visual Processing Mechanism With Multiple Pathways

As the increasing number of motor vehicles and influence of natural weather, the traffic environment is becoming more and more complex. In order to improve the road traffic safety, a set of vehicle detection system for complex traffic environment has attracted attention of scientists in the world. In this paper a moving vehicle detection algorithm based on multiple pathways visual processing mechanism is proposed, in which the multi-pathways visual processing and attention mechanism are inspired by the biological visual system. Target detection problem under complex environment can be resolved by this system.The moving vehicle detection system based on multi-pathways visual processing mechanism contains three major parts. The first part is the extracting of vehicles hypothesis area. Under the complex traffic environment, the moving vehicles in various orientations contain different information. The overtaking vehicles or passed vehicles on the right and left sides have obvious movement features. The motion information pathway of vehicles is constructed to extract vehicle candidate area of the left and right sides. The shape characteristic of front moving vehicles is symmetry. Meanwhile, every vehicle has a specific color feature. So in this paper, the candidate areas of vehicles are extracted by constructed shape and color information pathway of the vehicle. The second part is the verification of vehicle hypothesis area. The experimental data is collected by vehicle traveling data recorder. The vehicles drive into the receptive field of vehicle traveling data recorder with different position and angle. The vehicles in different visual orientation have different features. After balancing between the real-time and reliability of the whole system, the vehicle traveling data is divided into three different orientations data subsets. The vehicles of three different orientations include the moving vehicles in the left and right of host vehicle and the vehicles in front of host vehicle. The multi-block local binary pattern features are extracted for each data subset. Finally, three more targeted vehicle detectors are trained to solve the problem of moving vehicle detection under complicated environment. The third part proposed a set of development schemas based on heterogeneous dual-core chips OMAPL138 (C6748 floating-point DSP+ ARM9) for driver assistance systems. The experimental results show that the multiple pathways visual processing mechanism, compared with the single pathway AdaBoost cascade classifier, not only can reduce the complexity and training time of classifier, but also can improve the detection rate of moving vehicle.