Environmental perception systems for autonomousvehicles are often built using heterogeneous technologies thatoperate in a sequential manner. In the task of object trackingin particular, where the classical detector-tracker interactionis a serial process, it is viable to break the design rule byintroducing information loops. This is especially feasible in atracker that operates in a prediction-update cycle. Trackingpredictions can steer object detection towards regions wherean object is anticipated and, in turn, tracking updates canbe improved by incorporating reinforced detections. In thispaper we propose a novel detector-tracker feedback loop forinformation exchange based on the spatio-temporal similarityof detections and tracklets. We reinforce pedestrian detectionsthat have weak confidence scores by matching their boundingboxes to estimated tracklets with high tracking confidence. Theproposed system has several compelling advantages: based ona positive feedback principle it extracts the maximum detectionand tracking information, while operating transparently andwith minimal computational load. In a controlled ablation studywe evaluate our feedback mechanism using the KITTI objecttracking dataset. We show that our system gains significantperformance increase over the baseline in both frame-by-framedetection and tracking quality.