The Research And Implementation Of Pedestrian Detection Algorithm In Intelligent Transportation System

In the field of intelligent transportation, pedestrian detection is an important and fundamental task. The complexity of the scene as well as the diversity of body shapes and gestures make the embedded device based pedestrian detection a difficult task in computer vision. With the development of intelligent transportation systems, pedestrian detection system has broader application prospects. Meanwhile, it puts forward higher requirements on the system’s volume, power consumption, cost and stability. The research of embedded device based pedestrian detection bears important value. In this paper, a pedestrian detection based on Adaboost algorithm is designed and is implemented on the TMS320DM642platform. The main content and contributions include:1According to the study of pedestrian detection theory, the author chose Adaboost algorithm which can be migrated to the TMS320DM642platform. The designed system harnesses basic Haar-like rectangular features and triangular features specially used in pedestrian detection. The eigenvalues are calculated using integral images. OpenCV is used in the simulation to implement the Adaboost algorithm. The cascade classifiers are obtained by training. Many sets of experiments are conducted and the results are satisfactory.2In this paper, TMS320DM642chip of TI Corporation is selected as the main chip according to the design requirements. The author did close study on the configuration of the TMS320DM642’s video port and chose a suitable video capture&output chip and designed the PCB as a video device.3The development of the software is processed under the DSP/BIOS integrated real-time operating system. The designed system is divided into three parts:image acquisition, image processing and image display. The pedestrian detection algorithm based on Adaboost is then migrated to the DM642platform and the code is optimized using CCS tools. In the tests, the DM642based pedestrian detection system is able to process10frames per second, which meets the real-time requirement.