A real-time moving objects detection and classification approach for static cameras

Moving objects recognition plays an important role in camera-only active safety systems and intelligent autonomous vehicles. For these applications, reliable detection performance is required; however, pedestrian detection is challenging due to their divergent dressing and action variety. Besides, real-time detection and recognition performance is also critical. This paper aims to optimize the pedestrian detection and recognition by combining both temporal-domain and spatial-domain methods. Accordingly, we first use Background Subtraction (BS) technique to detect moving objects. Then, we use AdaBoost algorithm to classify the detected moving objects into their categories. Experimental results on our datasets show that the proposed approach can speed up 3.3 times in terms of processing rate, with significantly improved detection performance, i.e., at least 17% detection rate increment and 38% false alarm decrement for daytime out-door applications.